Home

Robotic Vision Platform

Contents

1. Be sure that back focal length is greater than 4mm KN CCD Camera C Mount Adapter Micro Video Lens Links Edmund Optics http www edmundoptics com onlinecatalog browse cfm categoryid 218 78 Appendix VIII Gearhead Analysis In light of a recent discovery that the specified gearhead is more expensive than anticipated additional gearhead options are provided The original specification of the gearhead was done under two assumptions One was that the budget was flexible and cost was not a significant factor in design and the second was that gearheads would not cost more than a 200 The specified Harmonic gearhead with a University discount is going to be almost 600 each While this is expensive if accurate object recognition are an integral part of the design this gearhead is absolutely necessary This gearhead has two orders of magnitude less backlash than all the gearheads considered It is one of the smaller and lighter options has the highest speed and can withstand the necessary torque for the pan and tilt If however object recognition is not planning to implemented with this design the more cost effective MicroMo and Moog gearheads will suffice dieta Drives Moog MicroMo 32 mm Precision a al 4 i CSF 08 30 2XH J Planetary Gearhead 30 1S 26 1S 26A Max Input Speed rpm 8500 5000 4000 4000 5000 Gear Ratio 30 29 23 23 40 Backlash deg 0 03 2 1 1 3 Weigh
2. Figure 34 Final Design without Shell Robotic Control System The controls for the robotic vision platform are performed by an independently operating PC 104 computer PC 104 is an embedded computer standard with a specific kind of connection that allows for a rugged stacking of boards Figure 33 The computer used to control the RVP runs on the Solaris operating system Solaris is a free open source Unix based Operating system On the computer are a number of PC 104 boards Figure 35 Top view of PC 104 stack showing PC 104 connection with white arrow 44 The PC 104 stack with a Solaris operating system was chosen because it duplicates the system used by the biped at IHMC By using this stack the controls can be tested at Bucknell or shipped to IHMC and tested there simply by replacing the motor control board from the top of the Bucknell PC 104 stack to that in the biped Solaris System Information Username root Password robot IP 192 168 1 50 Host 192 168 1 49 RTD Control Board The PC 104 board that is used to control the DC brushed motors on the head is a Real Time Devices ESC629ER 2 channel DC servo motor controller board The RTD board is capable of independently controlling two DC brushed motors as servo motors with encoders Based on feedback from the encoders and using user specified gains the controller is capable of controlling acceleration current velocity or position of the motors For the robotic vision pl
3. 61 this unexpected weight could have been avoided Finally in practice there only needs to be 2 cameras unlike the currently mounted 4 cameras The Bumblebee2 camera is a substantial portion of the weight at 75lbs If this camera system could be eliminated it would almost bring the RVP down to its specified weight However a main consideration in this design was robustness and versatility A platform which could test multiple camera sets was ideal so if the Bumblebee is not required the weight requirements would be much easier to make The Fire l cameras in the final design were selected in order to get the system up and running before the date of the design exposition More time should be put into selecting cameras It is thought that the only problem with the Firefly MVs which were used previously was that their signal output was not compatible with the stereoscopic multiplexor There may be other cameras which may be smaller fit better and give a wider field of view with interchangeable lenses More research should be done in order to select appropriate cameras Other additions for future work could be the addition of sound both a microphone for the user to hear the surroundings and a speaker for the user to communicate with them The Bumblebee2 if used should be incorporated somehow possibly for object tracking Right now it is not being used at all so someone should use it appropriately or not use it at all The program
4. en 36 Electrical Coppn ctlong i ianari eiaa aeaaaee aaia nad anaia aaia 36 Pan Mechita deed 37 Hard Stop and SENSO A ears eee a 39 BaS E 39 FireWire Hb Le lA 40 COMELACAOOIES eege AAA A AAA AA AR da 40 Robotic Control System ef nieret ane naani aeee a Rda 44 RTD Control Board ees ergeet che cdectecsandeessh aa 45 Compiling and Running the Software 48 Explanation Of Control ii A a AAE Ka aa a hei Aana 48 Spring Semester Programming Jesues 48 Overview Of Projects a 49 Overview of Primary Classes 49 Statement OfiEthiCS seh cc Gra A AA aie eel aerate 50 Statement of Analysis Of Safer 51 Statement of Sustainability ccccessssecececessesesseaeeeeeesseesesaeaeeeeeesseeseaaeaeeeeeesseeeaaeaeeeeseeseesseaaas 54 Testing Heeler EE 54 Testing of Previous Diese ia eege d Zeie deefe EES 54 Sp cification Testi dea ek ad ee sia Maedahaed oles edd aah a a a Aa E AEEA E Sie 55 Syste A NO 57 Results ON 57 Specification Result vc O da a 57 e EAR EE 61 FULULS Work A EE edu et eet 61 LESSONS Learned 0 cece ceccccccccccesessecceccececeueuceceeeeeecueuseceeeeueeuueuueecesesecauueeseceeeeseeuueuueseeeeuessuaueeceeeeeauens 62 AllOEAtION Gei 64 tele Tue le 65 Appendix I Stereo Vision Camera and Goggle Research 65 Appendix II Vision Platform Reouirementz 68 Appendix Ill Excerpt from Design of a Bipedal Walking Robot 70 Appendix IV Spechfications corn nnnnnnnnnnnonnnnnnnnnnannns 71 Appendix V Engineering Drawings ccccononoonon
5. 342g 2 x M12 microlens mount 157 x 36x 47 4 mm http www ptgrey com products bumblebee2 index asp Point Grey 2 000 Surveyor Stereo Vision System Headers for 8 servos 512x384 at 43fps 48 disparity GPL Open Source basic processing features WiFi through antennae Exposed circuit boards Built in dual motor driver 1A per motor http www surveyor com stereo Surveyor Corporation 550 PCI 104 nDepth Vision System PCI 104 camera lenses cables software 6cm baseline 752x480 stereo vision camera 30fps 92 disparity 1 3 wide VGA CMOS digital image sensors 4 25 x 1 5 x 1 25 in G2 system creates 3D object maps http www focusrobotics com docs focus_ndepth_pc i_brief pdf FOCUS robotics 3 995 DeepSea Stereo Cameras Used on Stanford Little Dog http www stanford edu class cs229 proj2007 Ki m GettingThePositionAndThePoseUsingStereoVision pdf Range 40 62 83 HFOV Aptina MT9V022 CMOS imagers 3cm 6cm 8cm 14cm 22cm 33cm baselines 512x480 at 200 fps 4 995 65 http www tyzx com PDFs Tyzx 20DS 20Cameras p df TYZX Video Goggles HMDs Name Description Price i glasses 920 3D Resolution 920 000 Pixels Per LCD Aspect Ratio 4 3 Color Depth 24 bit color Field of View 35 degrees diagonal Video Input Composite A V Video Input Format NTSC PAL SECAM 3D Video Format Interlaced 3D Video Audio Do
6. 8574K24 Polycarbonate Sheet 1 16 Thick 12 X 12 Clear 8975K413 3 8 Aluminum Plate 26955A25 HSS Hand Tap Plug 5 40 2 Flute que ome omo ome ome ome ome SS Ss ss gess ss sz 82 ome omo oma omo oms oms oms oms ome oms oms oms S Figure 26 BOM Part 4 Discussion During the process of fabrication various changes were made to the design Electrical components were much larger than expected and not enough space was allocated for electrical equipment Additionally the pan mechanism was changed to accommodate the wires and prevent them from twisting Finally new cameras were purchased after difficulties with Point Grey cameras and software Motor Gearhead Connector When the gearheads arrived in the lab it was found that the dimensions shown online were not entirely accurate There was an extra protrusion along the flange that did not allow the original motor to gearhead connector to fit properly The connection piece was made larger to accommodate the extra protrusion and allow the motor and gearhead to screw in properly Tilt Shaft Connection Originally the tilt shaft was to be press fit into the tilt bracket Later it was determined that friction would be insufficient to keep the shaft from slipping against the bracket Instead the 35 bracket was slotted with a screw hole going through the slot The shaft fit into the hole in the bracket and a screw was used to tighte
7. CS mount 5mm C mount adapter included e M12 microlense mount2 Video Data Output 8 and 16 bit digital data Image Data Formats Y8 Y16 monochrome 8 bit and 16 bit raw Bayer data color models Possible Options for 3D e Purchase nVidia 5 6 or 7 series card with 3D drivers unsure if newer cards support 3D output o Instructions in Z800 user manual support from eMagin o Few more options adjustments o Use 91 3 Drivers make sure forceware and stereo drivers are same 75 e Try using 3 party drivers from iZ3D with current ATI card http www iz3d com licenses o Talk to Charlie o Very little info or support base o Set s3D Mark to enable 3D o Need Direct X or OpenGL signal to activate 3D e Use Stereoscopic player multiplexer software from 3dtv at o Combine multiple camera signals to one file o Play stereo video files to multiple output formats e Talk to Virtual Realities about an integrated system Armand ext 449 x Quest 3D development tool for real time 3D apps x TriDef Media Player x Buy or upgrade to Dual input version 2195 plus signal conversion box Current Lab Set up 2D works 3D demo effect active but not synced timed correctly e Z800 goggles as second monitor from ATI card output 2D o Use iZ3D driver 3D o Force 3D using Lab Tool from eMagin Home Set Up 2D works 3D demo works 3D works automatically in Command and Conquer Generals Age of Empires 3 3D activates in Battlefield 1942 amp 2 but computer
8. Mertz uses YARP an open source vision code software with various libraries 3 Communication hardware and protocols a Interface with biped 4 Situational awareness a vision for tele operation b obstacle detection c location 5 Human operator a Work with Human Robot Team navigation software Basic 1 Fit on biped 2 transmit video to Carff system 3 provide pan tilt and torsion Performance no bigger than human head height 7 4 inches width 5 inches eye distance 2 4 inches Weight lt 4 25 lbs Mertz weight that Pan 90 degrees Tilt 90 60 degrees Torsion 10 deg Speed 180 deg sec Resolution 640 x 480 24 bit color nominal Capture speed need no more than 30 frames sec Isolation small picture movement EI ele ES ON E Wow 1 Binocular Controllable vergence 3 Cool looking head D 69 Appendix III Excerpt from Design of a Bipedal Walking Robot Design of a bipedal walking robot Jerry Pratta Ben Kruppb alnstitute of Human and Machine Cognition 40 South Alcaniz Street Pensacola FL USA 32502 vYobotics Inc 2138 Sinton Avenue Cincinnati OH USA 45206 ABSTRACT We present the mechanical design of a bipedal walking robot named M2V2 as well as control strategies to be implemented for walking and balance recovery M2V2 has 12 actuated degrees of freedom in the lower body three at each hip one at each knee and two at each ankle Each degree of freedom is powered by a force contro
9. 15 EPOX 635312 635 Thin Epoxy Resin System Quart 100z U S Composites Quart 211 1 20 75 16 EPX P11 Resin Pumps 1 1 Ratio U S Composites Set of 2 6 1 6 25 17 VB P56150 Peel Ply Release Fabric 60 wide U S Composites Yard 6 1 5 50 18 PD E16 Epoxy Parfilm Mold Release Spray U S Composites Can 13 1 12 95 19 KLY 01 Klean Klay U S Composites Lb 3 2 5 00 20 REX 224 Baal 2 Paste Wax U S Composites 24 tub 10 1 9 75 21 SQ 04PK 4 wide Squeeges Spreaders U S Composites Pack 25 10 1 10 00 22 SHR KV023 3 Blade Kevlar Shears U S Composites 28 i 27 50 23 GLV TD050 Tongue Depressor Mixer Sticks U S Composites Pack 50 2 1 2 25 24 GLV LL100 Disposable Latex Exam Gloves U S Composites Pack 50 10 10 00 25 VB BLE060 Bleeder Breather Cloth U S Composites Yd 4 1 4 00 26 VB BT25 Sealant Tape U S Composites 25ft roll 7 1 6 95 27 VB VF02110 Nylon Baggin Film U S Composites Yd 4 3 12 75 Figure 23 BOM Part 1 33 Super High Gain Micro Audio System 19 98 1 8 Miniature Speaker Visaton BF45 4ohm 17 39 1 1 Miniature Waterproof Speaker 6 62 3D Visor Z800 1 314 00 Firefly MVFFMV 03M2M C 400 00 Bumblebee 2 2 000 00 Unibrain Fire I Digital Camera 218 00 Firewire Hub 139 00 6 4mm FL High Res Video lens 78 00 C Mount to u Video Lens Adapter w Oring 50 00 M12 Lock Nut for Micro Video Lenses 15 00 Joystick Saitek X52 Flight Control System 89
10. an angle newAngle using the setPosition method in the ESC629ERAxisController class For additional information on the Java programming language a useful reference is the tutorial on the Sun website http java sun com docs books tutorial For more specific questions on the various built in classes and methods simply go to the first website found by Google by searching the name of the class or method followed by java This should bring up an in depth explanation from Sun Microsystems The first step was to write a driver for the Real Time Devices ESC629ER Motion Control Board This driver manages binary signals to and from the board by utilizing methods that simplify reading and writing data bytes to the two LM629 control chips in the board Our first board had to be sent back eventually because it had a defect that was giving us a great deal of trouble in our programming Most of the difficulty in writing the driver was the way that data bytes must be written and read from the LM629 control chips The chips must first be switched into COMMAND mode sent a command byte switched back into DATA mode and then the data bytes can be read or written After sending a command byte of sending receiving a second byte of data a bit in the status byte is set to logic HIGH and a method must be implemented to wait on this bit to return to logic LOW before continuing to read or write to the LM629 chips Another class was made called ESC
11. as much as the other design Figure 12 Neck Joint Concept 1 The second neck design Figure 13 was ultimately rejected for a couple of reasons First this neck does not have the parallel drive benefit that the first neck concept has This would significantly increase moment of inertia and weight of the motors Since weight is a driving factor for the design this was a strong reason not to pursue the concept any further Additionally there is an issue with the angle of vision Ideally the cameras should be able to 20 look around their range of motion while maintaining an image that is parallel to the ground However the second concept performs tilt before pan if it were to tilt down and then pan to the side there would be unwanted rotation of the image Figure 13 Neck Joint Concept 2 A major disadvantage of the neck concept is that the moment of inertia is greatly increased by moving the cameras off of the rotational axes This would increase the power required to move the head and call for larger motors It needs to be determined if the increased size of the motors to account for the larger moment of inertia would increase the overall weight of the head beyond specifications Another disadvantage is that the neck has a limited range of motion When the tilt axis is placed below the pan axis the camera can only look up or down directly in front of and behind the torso it cannot look sideways and down The current design cannot move
12. base In order to remove it one simply needed to unscrew a couple of bolts In previous designs such as in the summer of 2009 the shell was difficult to detach which maintenance time consuming and frustrating In the final design there were two simple plastic coverings over the cameras which spanned 150 degrees so the cameras could tilt and still see outside the shell This covering is not spherical and therefore could not distort the image Figure 20 Figure 20 Final Shell Design The shell was also made of fiber reinforced plastic and designed as small as possible to decrease weight and size In order to make the shell two molds were constructed one for the back half and one for the front half Figure 21 These molds were created in the rapid prototyping FDM machine They were then covered in epoxy resin and then sanded to get a smooth inside finish Wax and a spray on lubricant were applied to the molds as a release agent Two layers of carbon fiber and Kevlar hybrid fabric were laid on top of each other at alternating orientations letting the first set before the second was applied A vacuum pump device and airtight bag were used to make sure the fabric conformed to the mold After fully drying they were machined to mount on the RVP pan base A SolidWorks model of both the molds and the shell can be found below 27 Figure 21 SolidWorks Model of Shell Molds 3D Vision Design Once the Z800 3D vision goggles were received the
13. drawing files can be found in Appendix V 150 pa THREADED HOLE FOR A STANDARD 10 32 ANS SCREW TYP SIZE TO PRESSFIT 3 0 BEARING 1 FOR BE 64 ml Fe DIMENSIONS At H WENES DRAWN MRR Gaam SH TE WAC ONA 24 ER EE e Pan Bracket Arm ennen Se EKER II Tea B Atl ERA As Date Eat Gi AL6061 16 SEE DWG NO REY HE H PATIO ASA e vn A E o mee mes NONE RVP O5 SES Promo DECH bo sonenn Spee SCALE 1 2 WEIGHT N A SHEET OF 1 5 4 3 2 1 Figure 16 Pan Bracket Arm Drawing Shell Design The design of the shell or outside covering of the head went through a number of concepts to find the best combination of low weight small size functionality aesthetics and manufacturability Initial designs attempted to cover the whole moving portion of the head in a single casing that would both pan and tilt One concept attempted to follow a relatively anthropomorphic shape that kept the surface of the covering close to the cameras to reduce size and increase field of vision Figure 17 However a large circular radius was required at the bottom of the shell to allow a full range of tilt motion thus making the design large and unsightly Another concept avoided the issue of the shell intersecting with the inside brackets by making the shell a complete sphere A clear plastic or glass shield placed on the front allows the cameras to see This design however required the whole shell to be as large as
14. if a custom graph is able to be carried back through the Stereo Player 41 3d video 1 GRF GraphEdit File Edit View Graph Favorites Options Help Deg ai lei incl e ei ol e sl D POltGrey Camera d Point Grey Camera 0001 PolitGreyCamera e d Point Grey Camera r Seal d voz non a0 m gt A S t SI oprt w i rec avi Map un Game Ab wm Mur Siareoscoplo Mitilexer mart Tee MUPEG Compressor 0001 A Prevew l wm DI 3 t at Stereo Transformation Fiter smmplee wn wn xFom sl d wo Si e D Color Space Converter 0014 MUPEG Compressor 0008 wees pet XFom Ost hp Onprt XFom It D Color Space Converter 07 MIPEG Compressor Mires T007 vu weg ideo Renderer Figure 32 Attempted GraphEdit After this the Flycapture software was uninstalled and an alternate PGR package called Flystream was installed After re associating the Firefly MV cameras with this driver the Multiplexer did begin to recognize separate instances of each camera when 2 or more were connected As a result the configuration did progress even allowing for a live test preview of both cameras However after finishing the last step of the configuration wizard and trying to actually run the cameras through the multiplexer the process failed and an error suggested the program could not connect to any Point Grey device The same error appeared in the Player Further rese
15. operate in an urban environment Figure 10 Since the torso and legs of the robot are already constructed the platform needs to be able to integrate with the rest of the biped Based on work conducted on the previous iterations our primary focus for this iteration is to minimize the backlash between the motor and gearhead as well as in the transmission of power from the motor to the bracket The vision of the human head is capable for rotating along three axes of rotation the pan motion of head shoulder to shoulder the tilt nodding of head up and down and the yaw twisting the head In order for the robot to have an acceptable range of vision it was determined that pan and tilt motion are necessary A third axis of rotation or the yaw motion however was determined to be unnecessary Additionally it has been difficult to determine the best way to obtain 3D video between the 3 camera systems Therefore the design is versatile and is able to implement multiple systems using a standard photographic screw attachment The design of the platform should be as light as possible in order to reduce the torque requirements and size of the motors on the head as well as the rest of the body Also there are several electronic components that will be placed on this platform so locations and paths for the wiring need to be considered during the design process Figure 10 IHMC Bipedal Robot 10 In order to facilitate and guide the development of t
16. the design to create a stereo image Figure 5 This would require the team to correctly align and calibrate the image which would only be used for 3D video output to an operator However the Firefly MV is small 1 7 in wide light 0 08 Ib and is able to see Infrared light if a filter is removed Figure 5 Bumblebee2 Camera Figure 6 Firefly MV Camera Unibrain s Fire i cameras were chosen in the final weeks of the project as a back up solution Figure 7 They have similar resolution 640x480 and form factor to Point Grey s Firefly MVs but are slightly larger 62 x 62 x 35 mm 60 g The team will use two cameras to create a stereo 3D image This camera however will work with the software to stream two live cameras simultaneously because of the different signal output The presence of 2 FireWire ports per camera is an additional benefit that allows them to be daisy chained together Figure 7 Fire i Camera Three Dimensional Display Systems In order to output a three dimensional image to a human operator multiple types of 3D displays were investigated The most widely used method to display 3D video is through the use of vision goggles that fit over the users head These use two small display screens to provide each eye with a slightly different image creating the appearance of a single image with depth Most models are expensive and offer a limited field of view Appendix The best combination of price and performance was
17. the full 90 in either direction in pan or tilt Design adjustments may eliminate this issue Evaluation Each of the design concepts has obvious advantages and disadvantages While the simple rotation design has a much greater range of motion and a low moment of inertia there are issues with the wiring mounting a helmet and one motor must be strong enough to move the other The neck design would have no major wiring issues would allow the head to see the feet and the parallel drives would reduce weight but there would be weight issues with the long moment arm and the range of motion would be limited Ultimately the simple rotation design was chosen After a discussion with the leading engineers at IHMC it was determined that the advantages of having a neck joint did not warrant the increased complexity The ability to see the feet from the head is not of substantial importance at this time If the operator does need to see the feet this issue will be resolved by placing a camera at the waist of the robot The wiring issues associated with the simple rotation design 21 will be addressed by implementing a hub so that all electronic equipment will be centralized Because the motors that were specified are very compact and lightweight the ability of one to move the other is not as much of a concern as originally theorized Thus the simple rotation design should be able to combat all of the major concerns and be able to meet all of the prev
18. wrapped around the pulleys after the head is complete they must be properly tensioned during assembly Big Gears The big gears transmission concept involves a small gear on the motor shaft driving a much larger gear on the camera mount This would result in a large gear reduction and the size of the big gear would determine how far away the motor axis could be placed from the axis of rotation The size of the gears however would increase the weight of the head Much like the worm gear transmission the big gears concept requires high precision for the machined parts Evaluation The decision came down to a few main factors which were placed into a datum design matrix Table 2 The worm gear was the datum or the standard by which the other forms of transmission were compared Some of the major concerns with the transmission system were weight backlash and difficulty of assembly 15 Table 2 Datum Analysis Big Small Series of Criteria Gear Gears Low Weight Low Backlash Size Ease of Assembly Range of Motion Complexity Machinability Adjustability Distance From AoR 0 0 0 0 0 0 0 0 0 0 Gear Reduction Transmission Losses SUM SEA Series elastic actuators After analyzing the datum decision matrix the four highest transmission systems were further compared 16 Table 3 In the conception of the design the primary conc
19. 600 pixels Diagonal FOV 40 degrees One important decision to make while choosing the camera lens is the type of lens The two primary choices are either a fixed focal length lens and a micro lens The former would fit on the C mount adapter and the latter would require an adapter However the micro lenses are much smaller and lighter and capable of producing high resolution These micro lenses are also capable of viewing a wider FOV When comparing larger lenses to the micro lens it is important to note the maximum image sensor size Lenses with larger sensor sizes give a FOV that is for sensors of that size This FOV would be considerably decreased for smaller image sensors One of the primary concerns with this kind of lens was how to tighten the threading for the lens To fix this issue there is an adapter available with an o ring A concern that still needs to be addressed is how to fit the microlens on the camera at the proper focal length 77 Possible lenses Image Product Cost Focal Angular Diameter Length Weight Length FOV ay Edmund Optics 295 8mm 35 30 5mm 27 0mm 48g ay e Compact Fixed Focal Length Lenses e P NT56 526 p Edmund Optics 75 6mm 45 14 0mm 14 1mm TECHSPEC Megapixel Finite Conjugate u Video Imaging Lenses NT58 201 Edmund Optics 38 6mm 44 15 0mm 15 3mm S S Infinite Conjugate u 8 g Video Imaging Lenses
20. 629ERAxisController that would create an object for one of the controller axes In this class there are methods to set up initial conditions for the motors such as homing and setting up initial parameters There are other methods to change parameters during operation such as desired position and velocity and interrupts This class also has a doControl method that updates the necessary trajectory parameters and is continuously run during robot operation To get the robot head working with the input devices joystick and or slider board an embedded main is run on the PC 104 computer and a separate simulation class is run from a secondary working computer The two computers are connected via TCP Ethernet Both input devices are connected to and detected by the working computer and this is where the Yobotics Simulation Construction Set GUI is run Data is constantly updated for all the YoVariables in real time and graphs can be created to show and compare variables YoVariables are a special type of variable implemented by the Yobotics Simulation Construction Set Classes were created called BucknellHeadOneSimulation and BucknellHeadOneEmbeddedMain to serve these purposes Another class BucknellHeadOneController is essentially a code representation of the head system It sets up the axes on the ESC629ER board by creating two ESC629ERAxisController objects one for each motor There is a BucknellHeadOneController
21. 7 Figure 29 Fabricated Pan Mechanism In addition to dealing with the wires the new pan mechanism had more structural integrity and had fewer moving parts The new design rotated around two angular bearings whereas the prior design used two regular ball bearings and a thrust bearing Figure 30 Figure 30 New Pan Mechanism 38 Previously the bearings were exposed and all the force from the head was placed on the shaft If turned upside down all the force would be placed on one simple retaining ring Also the shaft was complex part to machine The new design incorporated the use of angular contact bearings which can take both radial and axial loads when preloaded correctly which eliminated the need for both thrust and ball bearings The inner and outer shells are used for support and are also used to correctly preload the angular contact bearings The main drawback is the tight tolerances required to correctly preload the angular contact bearings Hard Stop and Sensor A hard stop and a photogate sensor were implemented into the design In the event that both the program and the sensor cannot stop the motors from turning uncontrollably a hard stop was added to avoid the destruction of the expensive equipment that is on this RVP This consisted of a screw that is placed at the ends of the range of motion for both the pan and tilt axes This screw physically prevents the RVP from rotating further than intended A photogate sensor w
22. 939964847 Shoulder Screw 1 4 Shoulder Diameter 1 78 988044006 Stainless Stee Threaded Rod 24 5 40 2 79 936154405 Screw 18 8 33 Head Sekt Cap 1 4 20 3 8 length Pack of 10 1 s0 940354569 18 8 88 Precision Hedx Socket Shoulder Screw 1 4 diam 1 2 1 10 24 Individual 2 81 90259A124 Thin Wall Insert 10 24 int thread Pack 1 82 905924004 Hex Nut Class 6 M2 4 mm pitch 1 6 mmm h Pack 1 83 906954025 hex nut m2 4 mm pitch 12 mm h Pack 1 84 911134007 Lock washer no 6 screw size 3 OD 01 01 thick Pack 1 85 918014111 316 flat head Phil machine screw M2 10 mm lenggth Pack 86 Bearings an 90104006 18 8 SS flat head phil machine screw m2 size 12 mm length Pack 12 9 socket head cap screw alloy steel m3 thread 5 8 Pack 92210A150 18 8 SS flat head sckt cap screw 6 32 5 8 length Pack 90255A152 alloy steel button head socket cap screw 6 32 7 8 Pack 91255A081 alloy steel button head socket cap screw 2 56 1 2 length Pack 92949A114 18 8 ss button head socket cap screw 4 40 7 8 length 948314001 serrated flange hex locknut 6 32 5 16 width 11 64 0 all height 93501A007 serrated belleville washer no 6 M3 5 screw size 24 OD 02 thick 26309 ND Hex Screw Driver 9mm 035 27507 ND Hex Screw Driver 7mm 027 8586K162 Plastic ABS 12x12x1 8 910994167 4 40 1 2 Flathead Screw 91794A117 4 40 Fillister head 18 8 SS Screw 1 1 4 1556A51 Screw 8560K171 Clear Cast Acrylic Sheet 060 Thick 12 X 12
23. 98 Firewire 800 Extension Cable 12 inches 16 95 Firewire 800 Extension Cable 24 inches 18 95 Firewire 800 400 Cable 9 6 6 inches 52 50 Firewire 800 400 Cable 9 6 12 inches 35 30 C Mount 10mm 36 00 Photo interrupter Sensor 15 93 Coreless DC Motor Optical Encoder 594 90 Encoder Cable 22 50 Harmonic Gearhead GR 50 Size 8 model CSF 2XH 1 200 00 Carbon Steel Rod 5 32 Diameter 14 13 Rotary Motion Shaft 1 4 OD 14 13 Carbon Steel rod 3 32 6 length 12L14 Carbon Steel 3 84 Figure 24 BOM Part 2 Stainless Steel Pan Shaft a 1 3 84 Stainless Steel Tilt Shaft s 1 3 84 Aluminum 6061 Tilt Bracket s 1 3 84 Aluminum 6061 Motor Mounts and Pan Bracket so 1 3 84 Aluminum 6061 Base Plate 2 Pan Brack Arms sj 1 3 84 4 1 3 92 a 1 2 44 s 1 3 92 Timing Pulley GT2 2mm Pitch Pitch Dia 1 003 w Fairloc hub 65 a 195 00 Timing Pulley GT2 2mm Pitch Pitch Dia 1 003 w hub Se au 22 80 zl 2 10 38 9 o 18 22 4 7 20 s 2 10 48 6 au 276 3 3 5 10 6 2 12 18 SA 4 14 96 8 2 16 50 Figure 25 BOM Part 3 34 n 5 Angular Contact Torque Tube Bearing Thrust bearing 4 n 57155K322 Ball Bearing 1 4 Shaft Diameter 6 73 6384K354 Ball Bearing 3 8 Shaft Diameter 2 74 984104113 Retaining Rings External Snap 1 4 1 75 920104118 M3 Smm Flathead Machine Screws 18 8 Stainless Steel 1 76 918014111 M2 10mm Flathead Machine Screws 316 8 2 77
24. D Bett Timing Belt y a Figure 22 Comparison of belt profiles from Gates Corporation These belts were chosen because they are typically very thin but strong A 6 mm belt width with a 1 003 diameter pulley should be more than satisfactory for this application This decision was based on the strength of the belt In depth calculations can be found in the Safety section The cameras were specified to be able to have a very accurate resolution of 61 The field of vision FOV is a function of the focal length of the two cameras The FOV is a fixed value on the Bumblebee2 of 97 The focal length of the Firefly MVs is dependent on the lenses The lenses that are specified can only achieved a 43 7 FOV In accordance with the specifications the final design was a 2 DOF gimbal that has 180 range of motion in the pan and 135 in the tilt The Bumblebee2 and two Firefly MVs are implemented in 32 the design Because the Bumblebee is a binocular camera the total number of cameras was 4 The Bumblebee2 has a frame rate of 48 fps and a maximum resolution of 648 x 488 The Firefly MVs are able to realize 61 fps at a maximum resolution of 752 x 480 Budget and Bill of Materials This project received a sizeable grant for naval research Because of this achieving the above specifications and in a timely fashion were valued more than cost Despite high prices the necessary equipment to explore 3D vision and ensure minimal backl
25. TZ robot vision head are far beyond the scope of this project at this time The IHMC biped only requires stabilization and vision from this design Perhaps in the future this type of vision platform will be necessary for the IHMC biped to interact appropriately in an urban environment Because the functions of this robot head are more involved than what this project hopes to attain the size and weight constraints of the MERTZ robot were referenced as maximum values Figure 1 MERTZ Robot Head Robonaut Robonaut was developed by NASA in conjunction with Defense Advanced Research Projects Agency or DARPA Figure 2 It is intended to be implemented in space missions This robot is more similar in function to the biped developed by IHMC One important distinction is that Robonaut is a wheeled robot similar to a rover and thus does not have many of the constraints required by a bipedal platform The biped has strict weight requirement and requires much more stabilization than a wheeled robot Many of the specifications for the cameras used by Robonaut were not realistic for the RVP due to their large size and weight However the robot s head is intended to provide telepresence to the operator and move much like a human head would Because of the similarities in function the research that the designers of Robonaut conducted for the motion of the head was helpful in the design of the IHMC robotic vision platform Figure 2 Robonaut Robotic Vis
26. Table of Contents Executive SUMMA EE 4 Introduction and Motivation ocoonocccnnoccnanoncnonnnononnnnnn cn non cnn nn nnnn nn anno nr nn nn nn nn crac nan nrnn rante cnn 4 Background Resear hisses d Seel geen geg A ele eren e dE A MET EE 4 RODIN Utica a 5 kg E Ne EE 5 Vision Platform Design 1 Kandler and Snyder sssssssssrnssssssssenresssssserernnssssssrrennnssssesrrennnsso 6 Vision Platform Design 2 Rittase and Sirot cccccononococoonnnnnnonononnnnnnnonnnannnnnnoncnncnnnnnnnnonannnnnos 6 Three Dimensional Vision cirio ida 7 Three Dimensional Display Systems sssnnnssessseseesnsssssserresrsssssernesssssserernesnssssreernesssssseeennessssene 9 Customer Requirements and Engineering Spechications 10 e E e TEE 11 Si le El EE 13 Concept Design and Evaluation no nnnnnnnnnnnnnr no nn nnnnnncnnnnr nn nro 14 Transmissi SV EE 14 MOT GE S a ENEE 15 Cable Drives ii ia 15 Belleg cesta pad Ea aaia E ER adaa G 15 BIE GAIS ia 15 EVA ON in ds 15 CONC Suicida ti ii asii 19 LOETSCH it dd ta ride 19 el elei Ee E e 19 Neck e a E 20 EVA E 21 Design rte te Mon tii a 22 Key Design DecisiONS cion iia ERAN SEAN SEA 23 SAM end te te do cdo 25 BDMISION DEN Ee ee ENEE dee dE ee 28 Statement of Analysis of Motor Specification conos 29 Component Selec ii dia 31 Budget and Bill of Material 33 DISCUSSION cts lid 35 Motor Gearhead Connector nn nn nanrnnnnannnnnnnninnn 35 A O 35 Electronic Implementation tun dra dE A tae
27. The vision platform should be small and lightweight to minimize power requirements and the speed is intended to mimic the speed of a human eye The most important specification at this point in the project is the accuracy of positioning and stabilization of the camera system which will be primarily dictated by the backlash or play in the transmission system The speed of the motors needs to be fast enough for the image to be actively stabilized The biped has significant movement in its torso while walking due to a lack of any actuation in the torso It would be nauseating for an operator to deal with such large translations in the camera so the motors will use feedback from the biped s on board gyroscope to help compensate for this movement The previously built vision platform by Kandler and Snyder was able to attain speeds up to 1200 1300 degrees per second This exceeds the speed of a human eye which is approximately 800 degrees per second The Kandler and Snyder platform was significantly smaller it only carried one camera but this design will aim for similar speeds It will need to be 13 at least fast enough to actively stabilize the image This number was intended to be refined when the previous designs could be tested to accurately determine their speeds However due to issues with setting up the PC 104 stack we were not able to do any significant testing of the previous iterations Accuracy of positioning is particularly impor
28. achine Cognition IHMC is developing a bipedal robot designed to operate in an urban environment Currently the biped consists of a torso with two human like legs A senior design team has been assembled to develop a robotic vision platform RVP for this bipedal robot in order to provide telepresence to an operator and with the capabilities for 3D mapping in the future Background Research A number of specifications and design considerations of the robot head have been influenced by existing robots Two of the robots researched were MERTZ and Robonaut From this research two vision platforms were developed during the summer of 2009 These platforms were designed in parallel one by Matthew Kandler and Daniel Snyder and the other by William Rittase and Henri Sirot These designs have been utilized to provide insight into the next iteration of the vision platform Continuing from these designs research was focused on three dimensional vision and mapping along with refining the actuation transmission system Robotics Industry MERTZ The MERTZ head was designed by The MIT Computer Science and Artificial Intelligence Laboratory for analyzing social behaviors Figure 1 It requires continuous unattended operation interacting with humans for extended periods of time By interacting with humans it can identify and correlate objects and people and observe people s habits MERTZ can even learn to dislike people that annoy it The functions of the MER
29. al lenses included a matching photo screw mount and was readily available Most importantly it clearly stated that it supported RGB24 and YUV 4 2 2 output although at a less than ideal 15 fps It may not be the best long term solution but was a good last minute substitute in late April Initial installation of the Unibrain cameras was tricky because the computer thought they were PGR devices and tried to assign them PGR drivers even after all PGR software was uninstalled The Unibrain software and drivers that were included on the CD could not install without hitting an error but a newer version Fire i 3 80 downloaded from the Unibrain installed successfully This installed a Fire i application that allows a user to view and change the settings of all connected cameras and associates the proper Unibrain Fire i driver with the cameras Two cameras independently connected to a single Firewire card show up properly in the multiplexer software and allow for a complete successful setup This carries through to the Stereo Player which allows for live viewing of both streams in 3D in the goggles Thus everything works as advertised The only complicated part is aligning the camera angles to create a similar image for each eye The success from this setup suggests that the 2 cameras could also be daisy chained together For a full description of how to set up the cameras for 3D vision see the user manual Figure 33 Final Design with Shell 43
30. arch suggested that the root of the problem lied with the incompatible formats that the Firefly MV output Y8 Y16 8 amp 16 bit raw and that the Multiplexer accepted YUV 4 2 2 RGB248 32 It seemed this problem could be addressed in GraphEdit and in preview screens but could not be handled in the Multiplexer in a way that allowed for live 3D output to the goggles The only near term foreseeable option was to find new cameras that output a format supported by the Multiplexer Meanwhile the Bumblebee2 stereo camera was largely ignored in the attempts to obtain 3D video in the goggles Although specifications suggest the camera can output YUV4 2 2 and RGB formats discussions with Point Grey indicated that the Bumblebee signal is in Byte Interleaved format which alternates between left and right camera signals A special program or code would need to be found or created to separate out the 2 cameras to use 3D vision in the current set up This was not achievable given team experience and time constraints However the Bumblebee is still present on the head and can be used with PGR Triclops software to give depth maps of the environment A new search for cameras showed hundreds of possible solutions a list of which can be found at http damien douxchamps net ieee1394 cameras The Unibrain Fire i was selected from the list because it was inexpensive 109 close to the same size as the Firefly MVs had 2 FireWire 42 ports needed no addition
31. as originally planned to be placed at the zero location for the pan and the tilt mechanism or when the head is facing straight forward This was to be used as a safety measure to double check the position of the bracket in conjunction with the optical encoder on the back of the motor If the sensor gets to a certain position it will disrupt the program or shut off the power This however was not implemented The RVP was becoming cluttered after all the FireWire cables from the cameras the circuit boards and other wires were attached There was limited space on both brackets to place sensors Instead of implementing this idea just a hard stop was added Base Plate The base plate was modified to fit three circuit boards 2 x 3 inches in size Figure 31 These hold the gyro power safety fuses and other electronic signal connections and converters The boards could not be placed underneath the plate near the motor because the electrical signals could be altered by the magnet in the motor 39 Figure 31 New Base Plate The plate was redesigned to have two pockets where circuit boards were placed vertically This allows for a more versatile arrangement of the board and easier access to the wires exiting the bottom of the pan mechanism The base plate was also extended in the back to accommodate the third and final circuit board holding the gyro The area under the plate in the back was not used because it was believed things placed here wou
32. ash were able to be purchased Also extra stock material additional belt sizes and bearings were purchased as a precaution Figure 23 Figure 24 Figure 25 Figure 26 An estimate of approximately 10 000 was spent on the project including the cameras and the operating system for the RVP Part Purchase Description Company Material oe Unit Price Amt Price 1 MM 32DX AT PC104 Board Diamond Systems Individual 0 1 0 00 2 HE104 DX 108 PC104 Board Power Supply Tri M Individual 299 1 299 00 3 ETC NANO 104 PC104 Board Basebaord ACCES VO Individual 229 1 229 00 4 CAB ETX ATX PC104 Board Power cable ACCES VO Individual 20 1 20 00 5 ETX PM1 4G PC104 Board 1 4 Ghz CPU ACCES LO Individual 749 1 749 00 ETX LFHSPWFAN PC104 Board Heat Spreader with Cooling Fan ACCES 1 0 Individual 26 1 26 00 7 CPF 8GB 11 PC104 Board 8GB Flash ACCES LO Individual 219 1 219 00 8 SODIMM 1GB PC104 Board Memory Module 1 GB ACCES LO Individual 99 1 99 00 9 MAG05 0300S050 PC104 Board Mag 3 5 IMEMSense Individual 669 1 669 00 10 AIB PC104 Board Analog Interface IMEMSense Individual 110 1 110 00 11 FG CF5549 Carbon Fiber Fabric U S Composites Yard 27 1 26 50 12 FG CF5702 Carbon Fiber Fabric U S Composites Yard 5 2 10 00 13 FG CK99150 Blue Carbon Kevlar Hybrid Fabric U S Composites Yard 39 1 38 50 14 FG CK99450 Orange Carbon Kevlar Hybrid Fabric U S Composites Yard 39 1 38 50
33. asured by hand once the RVP is made before then we can refer to the CAD dimensions Weight 4 pounds Justification The average human head weighs around 10 pounds However this seems very high for a robot made of aluminum and cameras We believe a Bumblebee2 and 2 Firefly MVs can be incorporated onto the head with aluminum supports and be around 4 pounds at most The Bumblebee weights 342g 75 lb and the Firefly MV weighs 37 6 each 082 Ib This gives us 3 086 Ib for the motors and the mounting Validation Technique The weight can be measured using a simple scale During the design process weight can be approximated using volume and material density in CAD files 71 DOF 2 Justification Pan and tilt are necessary for the operator to be able to look around the environment comfortably During operation the biped does not roll very much so extra actuation is not required for this DOF Validation Technique This can be measured by counting the number of actuators in the design Speed TBD Justification We believe this value needs to be around 1200 1300 deg s as this is what the spec sheets for the motors on the IHMC head call for These motors appeared to move quick enough for active stabilization of the image it does not need to move any faster than this speed Any faster would be overkill However we are not sure if the specifications match what the actual design is doing so a test will be done later to determi
34. atform there is a motor to control pan and another to control tilt connected to channels 1 and 2 respectively Motor power can be supplied by an on board power supply on the RTD board either 5 or 12 V or by an external power supply For the RVP external power is used because the motors used in the RVP require more current than is available on the board and it is desirable to separate the power controller the motors and operation of the PC 104 stack A standard DC power supply unit is set to 12V and connected to the M1 V and GND screw terminals in order to provide the necessary power Figure 34 These terminals are wired to the M2 V and GND terminals for the second channel The purpose of this external power supply is to provide dirty power to be sure that the draw from the motors does not interfere with the power to the encoders or the gyroscope Motor wires are connected to the and terminals corresponding to the correct channel M1 or M2 45 19 ER Figure 36 Top view of RTD ESC629ER control board displays screw terminals on left For each motor channel there are five screw terminals corresponding to encoder power 5V and GND channels A and B CH A and CH B and INDEX The encoders are power by an on board DV power supply what is considered the clean power in the system Programming for the camera stabilization system was originally done in Java using Borland JBuilder 2006 so
35. attached Appendix IV Concept Design and Evaluation Design of the RVP was broken down into two major phases The first phase was to determine the type of transmission that would be used to drive the pan and tilt of the head and the second phase was to find the best means of implementing the transmission system with the least amount of play Transmission System Since pan and tilt were determined to be the only two necessary degrees of freedom the first design aspect was determining how to actuate the head in each of these axes A variety of transmission systems were explored to find which would be most appropriate for the head based on the design specifications After creating rough design concepts to go with each of the proposed transmission systems comparing characteristics using a datum decision matrix and thorough discussion a transmission system was chosen 14 Worm Gears One of the first options explored was worm gears A major benefit is that worm gears are not backdriveable This would also save energy because the motors would not have to work to hold the cameras at a specified position Additionally the worm gears are capable of producing large gear reductions This would eliminate the need for a gearhead and reduce weight and size of the head The backlash on worm gears can be as small as 1 2 arc min sec approximately 017 degrees which is well within our specification of 1 degree Unfortunately in order for such low value
36. blebee2 0 75 343 0 7 82E 05 1 66E 05 Firefly x2 0 17 75 2 1 42E 05 1 42E 05 Upper Pulley Tilt 0 01 4 5 2 95E 05 3 41E 03 Lower Pulley Tilt 0 01 4 5 2 65E 05 Tilt Shaft 0 02 8 5 7 10E 04 5 83E 01 Tilt Bracket 0 16 74 1 2 46E 05 2 38E 04 Bearing 0 00 0 8 7 12E 03 1 43E 01 Shoulder bolt 0 01 6 2 6 39E 04 4 74E 01 Pan Assembly 0 47 213 0 1 34E 06 Motor mount 0 07 30 7 4 59E 04 Motor 0 24 111 0 4 43E 04 Helmet 1 00 454 5 3 13E 06 3 13E 06 Gearhead 0 24 111 0 3 51E 05 Pan Pulley 0 09 40 7 3 32E 03 Pan Motor Pulley 0 09 40 7 1 44E 05 Total Excel 3 34 1518 6 6 93E 06 3 46E 06 2 The final values of the moment of inertia were roughly 6 930 000 g mm in the pan direction and 3 460 000 g mm in the tilt Final Answer Once the moments of inertia were summed together the final torque values were tabulated Table 6 The tilt motor was found to require roughly 109 mNm of torque while the torque necessary to move the RVP in the pan direction was about 218 mNm 30 Table 6 Final torque values Req d Torque mNm Motor Torque mNm FS Pan 217 82 800 3 67 Tilt 108 82 800 7 35 Due to the size constraints the motor required needed to be extremely small Also since the motor had to move quickly and in both directions a DC motor was chosen A motor was specified from MircoMo one of few companies that manufactured extremely small yet powerful motors The chosen moto
37. cnnnnnnnnononnnnnnnnnonanononnnnnnnnnonanononnnnnnnnnnnnnrnnnnnnnnanannnos 74 Appendix VI 3D Stereoscopic Vision Components 75 Appendix VII Camera Lens Selecton nono nnnnnnnnnnnrnnnnnnnnnnannnns 77 Appendix VIII Gearhead Analysis ccccscccccccecsssesssececeesceeseseaeeeeeceseeseaaeseeeeseesseaaeaeeeeeeesees 79 Executive Summary This report describes the details of a robotic vision platform to be implemented on a bipedal robot designed by the Institute of Machine Human Cognition IHMC This design is the third iteration of a vision platform that is intended to provide image stabilization during operation as well as telepresence with the potential for object recognition The design will explore 3D vision and significantly decrease backlash when compared with the previous designs Vision in 3D will be implemented with the addition of 3D goggles as well as testing two methods of achieving stereovision Two different methods of obtaining 3D vision will be implemented in the design a binocular camera system Bumblebee2 and two discrete cameras Fire i The platform is a 2 degree of freedom system controlled by two geared DC brushed motors to provide sufficient stabilization Two sets of synchronous belts and pulleys are implemented to transmit power This vision platform is light and accurate and will enhance the functionality of the biped Introduction and Motivation Bucknell University in conjunction with the Institute for Human and M
38. considered Cameras It was determined that both the Bumblebee2 and the Firefly MV cameras would be used on the robotic vision platform This would allow the RVP to have a degree of modularity through the use of standard photographic mounting screws The Bumblebee2 could be used to perform 3D mapping and the Fireflies could be used for telepresence Ideally the Bumblebee2 could be used for all the desired functions of the head but there were concerns with how much the camera could handle and how the data would be split up to perform multiple functions Additionally the Fireflies can be placed at a distance that models the human eyes and results in a more natural feeling telepresence Both cameras will be designed into the RVP but in a way that they could be removed if not needed This will provide the option of reducing weight once the capabilities have been fully tested Simple Rotation Simplicity is often times the best approach The simple rotation concept has a pan bracket and a tilt bracket that are both being rotated about the axis of vision Figure 11 This is preferred because the cameras will not be translating at all during motion just rotating Figure 11 Simple Rotation with both Bumlebee2 and Firefly MVs Due to the simplicity of the design there is also a large range of motion allowed by this design The camera will easily be able to move 135 in the tilt and a full 180 in pan By rotating the 19 head about the vision a
39. deg down from the horizontal should be adequate Validation Technique The range of motion may be limited by either the geometry of the head and some sort of hard stop or by software limitations Once the design is finalized the ranges of motion can be measured by checking the limits on pan and tilt Number of Cameras 4 Justification In previous meetings the group has come to the decision to include the Bumblebee2 with two cameras on it and two of the Firefly MVs This may change depending on interfacing between the PC104 and the cameras themselves Validation Technique This is measured by counting the number of individual cameras on the final design Camera Frame Rate 30 fps Justification Current cameras and previous specifications call for 30 frames per second as minimum This number is also what the human eye can interpret Validation Technique This is a characteristic of whatever cameras are chosen to be used on the final design that can be obtained from specification sheets Field of Vision 60 deg Justification This number is based on the average FOV of the 3D vision goggles that we have been looking to purchase We will resolve this further once we physically can subjectively test them in the lab Validation Technique This is a characteristic of the type of camera and lenses that are used The 3 D goggles will likely have a smaller field of vision and limit the system To check the accuracy of the FOV from sp
40. easier to organize but working on two or three group design projects simultaneously made meetings extremely difficult to organize Being flexible and adjusting to the needs of our teammates was integral in our success As long as we went about scheduling meeting times in a cordial manner things 62 never got heated The buzzwords for good teamwork especially in a senior design team are flexibility and respect Though we have worked on many projects at Bucknell senior design has been our first project that has spanned an entire semester Because of this we have had to learn to allocate time to meet frequent deadlines One of the biggest challenges with meeting deadlines was finding a way for the entire group to meet Sometimes we had to meet at inconvenient times or only for a short while These short work periods were usually not very productive The bulk of our work was done during longer work periods where we felt like we had enough time to dive more deeply into the work In general we tried to find a way to keep everyone updated rather than just breaking up the tasks and having each member only aware of their own direct work Sometimes this would cause us to slow down on decision making From this we learned that it was important to use our individual time more effectively and trust and rely on the other members of the group to complete their tasks In order to get tasks accomplished we found that we do not all need to meet but sometimes brea
41. ecifications we could use simple trigonometry with the camera images Resolution 640 x 480 Justification The Bumblebee and Firefly MVs typically give off a 640x480 resolution image This can be increased but it lowers the frame rate and requires more processing power Validation Technique This is also a characteristic of whatever cameras are chosen to be used on the final design that can be obtained from specification sheets 73 Appendix V Engineering Drawings 74 Appendix VI 3D Stereoscopic Vision Components Hardware e eMagin Z800 3D goggles http www 3dvisor com support ph o Requires frame sequential video input through VGA in DirectX or OpenGL o 800x600 60 Hz signal e Dell Optiplex 960 Intel Core Duo E8600 3 33GHz 3 25GB RAM Windows XP SP3 o ATI Radeon HD 4670 graphics card with 512MB memory Driver 8 561 0 0 12 1 08 e Point Grey Bumblebee Stereo Vision system http www ptgrey com products bumblebee2 o BB2 03S2C Sony ICX424 CCD 648x488 pixels 48 fps o BB2 0852C Sony ICX204 CCD 1032x776 pixels 20 fps o Image Formats On camera conversion to YUV411 YUV422 and RGB formats o Focal Lengths 2 5mm with 97 HFOV BB2 only 3 8mm with 66 HFOV 6mm with 43 HFOV e Point Grey Firefly MV cameras http www ptgrey com products fireflymv index as o 2x FFMC 03M2C Micron MT9V022 CMOS 752x480 pixels 61 fps FireWire o FMVU 03MTC Micron MT9V022 CMOS 752x480 pixels 61 fps Mini B USB o Lenses
42. ed and placed on a backpack The head was placed on the wooden frame and secured One of the members of the team will walk around with the backpack and the gyro will be tested to see how well it can compensate for this walking motion The gyro on the head reads information as the head is tilted and sends this information back to the PC 104 This will cause the cameras to compensate for the motion of the head and this will be measured based on the output of the cameras on the head The cameras need to compensate for side to side yaw motion of the walking robot as well as up and down pitch motion The design did not account for roll or twisting motion and thus this will not included in the testing of the device Results Specification Results The Robotic Vision Platform was tested against the specifications These tests needed to be performed without the helmet because the helmet was still in the process of being fabricated The addition of the helmet would have only altered the dimensions weight and dynamic characteristics slightly The results for each specification test were tabulated Table 13 57 Table 13 Testing Results insufficient results in red BE Tested Value Specifications Value w o Shell Size smaller than 8 x8 x 7 x9 x8 5 10 DxWxH Weight 4 pounds 4 82 lb DOF 2 2 Min Speed 200 rpm 94 5 rpm Min Acceleration 1800 deg s 6916 deg s Accuracy of Less than 1 N A P
43. ee dimensional robot2 We believe that high fidelity force control is a critical requirement for graceful walking over rough terrain and robustness to disturbances Our previous work with the robot Spring Flamingo1 10 11 has shown that good force control leads to simple yet reliable algorithms for walking over rolling terrain Figure 2 shows time lapse photos of Spring Flamingo in 1999 walking over alternating 15 degree inclines and declines without any prior information or sensing of the terrain besides the location of where its feet fall To date we know of no other bipedal walking robot that has been able to repeat this feat We credit Spring Flamingo s graceful walking in part to the force controllable Series Elastic Actuators33 at its joints An improved version of these actuators are used in M2V2 70 Appendix IV Specifications Size smaller than 8in x 8in x 10in DxWxH Justification These dimensions are based on proportions of the human body and head from Da Vinci s Vitruvian Man These dimensions were then checked against a typical human Also the width dimension needed to be able to fit the Bumblebee so it was increased slightly to account for a helmet Additionally the space needs to accommodate at the minimum all 4 cameras We plan to use 1 Bumblebee 2 47 4x157x36 mm DxWxH and two Firefly MVs 24 4x 44x34 mm DxWxH each The remaining space is shown below by the water box Validation Technique Dimensions can be me
44. ern was the amount of backlash that the transmission would introduce into the system Thus the table assessed backlash torque typical applications and a summary of other notable qualities 17 Table 3 Transmission Analysis Torque Backlash Applications Summary High gear reduction Differential H SN f eavy Range 1 2 systems in cars R 3 high Si Worm limited to 750 to 28 48 milling operations equires Ne poner Gear hp arc toys and small Noisy at high speeds minutes electrical bulky operations Moves actuation away from axis ae to ba ihan Silicon wafer of rotation AA E Wrap saws positioning Eliminates backlash cable drives Cable SR units for Work equally as well as speed For multiple minimal Zen Drives directional solar increasers or decreasers cable drives i d rob range to panel and ropot Requires tension 76 000 Ib ft ii seint flexible Requires tension Thin flexible allows them to limited to 200 operate well on miniature hp Power saws drives and in apps requiring high motorcycles speeds or small pulleys Belt Small ee vacuum cleaner o minimal Maybe the most efficient form Drive maximum gear brushes printers WE ratio about l sar cutter of power transmission short of 10 1 direct drive Can sustain high loads Low precision required Easy calculations Requires high precision Anti Highest torque capability Clocks and backlash f High gear reduction Big watches washing Reall
45. ers the current and thus the temperature increase were calculated Table 11 Current through motor and temperature increase Current 0 059913 amps Tinc 0 458745 C It was found that the increase in temperature operating at normal conditions would be less than 5 C or about an increase of 9 F This was said to be an acceptable increase in temperature This should not cause any damage to the electrical equipment 53 Statement of Sustainability This particular project was not focused on sustainability however it was considered in part in the design process Most of the components of the RVP were made from aluminum which is easily recyclable and is easier and less invasive to mine than other metals Additionally material use was minimized Aluminum was chosen to be as thin as possible while still maintaining the structural integrity of the RVP The material that was used in construction of the RVP was often taken from scrap material already in the Product Development Lab at Bucknell Sustainability was also considered in developing the testing apparatus A recycled backpack from a former senior design project was modified instead of purchasing a new backpack and thus wasting more material Extra molding from Professor Shooter s house was mounted onto the backpack to create a stable platform to support the RVP Testing Procedures Testing of Previous Designs A list of testing procedures to
46. found in eMagin s Z800 stereo vision goggles Figure 8 These lightweight 8 oz goggles have high resolution 800x600 OLED displays a 40 field of view built in 3 axis head tracking for movement of the RVP corresponding to the operator s head position microphone and speakers and relatively wide compatibility with graphics cards They require frame sequential video input via a single VGA cable These goggles were ordered for 1299 Figure 8 eMagin Z800 Goggles Figure 9 iZ3D Monitor Another option to display a 3D image is the use of special 3D computer monitors or TVs Figure 9 Although the technology is less well known it has been around for about 5 years and is finding increasing market use Most versions require users to wear either passive polarized glasses or active shutter glasses while some screens need no eyewear but do require observers to be in specific viewing positions The most promising setup is made by iZ3D Technology its 22 9 LCD monitor has a second polarizing screen overlaying the normal LCD thus creating an image that is seen in 3D when the user wears passive polarized glasses The 300 iZ3D monitor is also compatible with the Stereoscopic player and could be considered as a way for a group of people to watch a 3D demo simultaneously Customer Requirements and Engineering Specifications As mentioned previously the goal of this project is to develop a robotic vision platform for a bipedal robot that will
47. ft had a pulley attached via set screws on a 1mm deep groove The other end of the shaft slides through two flanged ball bearings on the pan bracket arm To axially constrain the shaft two snap rings were placed in circular grooves on either side of the flanged bearings Pan Shaft The pan shaft was attached to the bottom of the base of the pan bracket by a hole and flange Alignment was achieved by fitting the top of the shaft into a hole and then it was held in place by four screws on the flange The pan shaft carries the load of the entire head and it was important that this shaft was properly constrained In order to handle both radial and axial loads a combination of ball bearings and a thrust bearing was used A support bracket was used to eliminate moment produced by loading from the pulley placed in the middle of the shaft Flanged ball bearings are used in this support shaft and base plate There was a shaft diameter reduction at the end of the shaft that allowed the shaft to rest on a thrust bearing to handle axial load To account for any axial load pulling up on the head a snap ring was placed beneath the flanged bearing in the base plate During fabrication it was determined to completely redesign this mechanism to accommodate wiring and electrical components Please see the Discussion section for more detail 24 The final design was drawn in SolidWorks An example drawing file is attached below Figure 16 and the remainder of the
48. ftware This later version of the software was important to use because it is the same version used by IHMC because of issues with some of the updates However this year IHMC switched to JBuilder 2008 and all appropriate changes were made to the code There are few obvious changes except that it takes considerably longer to send the files from the host computer to the PC 104 computer Java is an object oriented programming language created by Sun Microsystems In Java script can be organized using files called classes A specific instance of a class is referred to as an object This could be compared to the way that Bucknell is a university Bucknell is a specific instance of the more general term university Each class can contain any number of variables and methods Methods are functions within a class that can perform a variety of tasks such as storing and retrieving variables The method setPosition for example sends a signal to the ESC629ER board that sets the desired position to a specified angle The following lines of code provide an example of how a class is created and a method can be called using the class ESC629ERAxisController ESC629ERAxisController panAxis new ESC629ERAxisController creates an object panAxis the is where required information is fed into the constructer for the class such as axis number and name panAxis setDesiredPosition newAngle 46 sets the desired position for the object panAxis to
49. g ratio and taking the derivative of the position data for both the pan and tilt axes The speed appeared to be limited by the range of motion allowed and a wider range might allow for higher speeds Acceleration Accelerations were determined using similar techniques as were used to determine speed however to obtain the acceleration the second derivation of position was taken Accuracy of Positioning Position the head to move just before the hard stop Gradually move the head towards the hard stop in increments of 5 Keep moving until the hard stop is hit and record distance travelled to get there This can be compared with the distance that the computer believes it has travelled to compare This was not completed yet Range of Motion Measure distance from hard stop to hard stop for each DOF 56 Camera Frame Rate Point Grey software will display frame rate This can then be compared to the specifications listed Field of Vision Turn on cameras Mark off in the room the locations that the very last edge that the camera can see Measure this distance to the center of the head and determine the radius From this the angle can be determined using the equation 2 tan W 2 D where D is the depth to the surface and W is the horizontal width System Testing Compensation for Walking It was important that the head be able to compensate for the motion of walking The head was sitting on an aluminum stand and a wooden frame was construct
50. h eventually gets the head stuck at the software limit in the pan axis Currently another gyroscope is being purchased to ensure that the current gyroscope is not just a lemon Further work needs to be performed to adjust the calibration properties which are currently set to those from the biped project files Because the problems are limited to the Z axis it seems likely that there is some issue with the compass integrated into the gyroscope More in depth analysis of the code and variable adjustment needs to be performed to identify this problem However when the head is eventually mounted on the biped it will not need this gyro and can run off the existing gyro This has already been proven to function properly during by testing done at IHMC with the previous iteration Overview of Projects RobotHeadControl Primary project file ESC629ERAxisController BucknellHeadOneController BucknellHeadOneEmbeddedMain BucknellHeadOneSimulation HeadInputDevices RobotConstructionSet Contains reference files RealTimeDevicesESC629ER SimulationContrustionSet Contains reference files SimulationConstructionSetUtilities Contains reference files Overview of Primary Classes RealTimeDevicesESC629ER This is the driver for the ESC629ER Motion Control board It handles all of the bytes to be sent or received by the board The driver is generalized so that it is compatible with an ESC629ER control board regardless of the specific motors be
51. haft mates with a receiver of matching size on the gearhead and is secured with 2 set screws A special connecting plate between the motor and gearhead hold the two pieces together and attach the powertrain to the head The mounts for both motors are slotted to allow for proper tensioning of the timing belts There is a flanged output on the gearhead so a special shaft will need to be machined to connect the gearheads to the pulleys The same motor gearhead pullley combination is used for both pan and tilt to simplify the design controls and machining The motion of the head was specified to achieve 200 rpm and 1800 deg s The DC motor that was specified has a no load speed of 8500 rpm After the gearbox this speed was reduced to 283 rpm which is still above the specified 200 rpm In the motor calculations an acceleration of 1800 deg s was used to determine motor torques to ensure that the accelerations would be adequate The motor was specified based on these calculations thus should be able to achieve 1800 deg s The calculations for these numbers are shown in more detail in the Statement of Analysis of Motor Specification section of this report GT2 belts were chosen to transmit the power from the motor to the robot head This type of timing belt was specified because they were specifically designed to minimize backlash and increase accuracy These belts have less backlash than typical GT and HTD belts Figure 22 y PowerGrip GT2 Belt HT
52. hanically and electronically robust mean time failures of several months Real time feedback Easy to write compile download and run software Easy to run and maintain able to operated by single person Robustness to disturbances 12 By identifying the potential scenarios that the robot might encounter the team was better equipped to specify the constraints on the design Specifications Given these requirements at the start of the project information from the previous designs and feedback from Dr Peter Neuhaus at IHMC the Robotic Vision Platform Team generated a list of engineering specifications Table 1 Table 1 List of Specifications Specifications Value Size smaller than 8 x8 x 10 DxWxH Weight 4 pounds DOF 2 Speed 200 rpm Acceleration 1800 deg s Accuracy of Positioning Less than 1 Encoder Resolution 01 Range of Motion 180 pan 150 tilt Number of Cameras 4 Camera Frame Rate 30 frames per second Field of Vision 60 Resolution 640 x 480 Most of the requirements defined by the IHMC researchers identify the specifications of cameras to be implemented in the design such as frame rate resolution etc The initial camera systems Bumblebee2 and Firefly MV were purchased using these specifications in the summer before the team was put together Therefore the team largely focused on the size weight and backlash of the actuation transmission systems
53. he RVP the project s advisor Professor Steven Shooter required certain deliverables through all stages of the design process The following is a list of all items to be completed by the end of the project Comprehensive review of web and other literature on the design and development of robot vision platforms Development of detailed design specifications Development of several conceptual designs of potential vision head configurations Theoretical and computational evaluation of vision platform concepts Selection of the concept and detail design of all components including drawings of all parts selection of sensors and specification of elastic components Purchase of vendor supplied components Fabrication of remaining components Assembly of the vision platforms Formulation of testing procedures and comprehensive testing of performance Refinement of design to improve performance User s manual video of performance of final design and recommendations for further improvements to the design Regular communications and updates to IHMC researchers Complete and detailed project reports at the conclusion of the fall and spring semesters Each of these requirements has been completed The team researched several heads from other robots and gained significant insight into the state of the art robot heads Design specifications have been intricately tuned by the tea
54. hem out there use 2 cameras b Control i Focus ii Aperture for variation in lighting iii Vergence if using two c Motions do we need eyes to move independent of the head i Pan 1 or 90 degrees Half of full spectrum 2 Speed 180 deg sec Other heads seemed able to go 150 deg sec 3 Incremental resolution 0 003 degrees TRISH paper Need to explore what is needed for isolation of body movements 4 Will need to compensate for 30 degrees of swing during walk to isolate image during body yaw ii Tilt 1 90 60 degrees Can look completely up but any lower would look at body Speed 180 deg sec to have same as pan Incremental resolution 0 003 degrees TRISH Balance at home position Will need to compensate for 10 deg of tilt from body during walk 6 Strongly desired to use tilt to compensate for 7 cm of z translation of the body during walk Will need to choose a focal distance for this Maybe 2 meters iii Torsion 1 10 degrees Need to account for body roll during walk 2 Speed 180 deg sec to keep the same Not really needed as fast would think OM ey eS 68 3 Incremental resolution 0 003 degrees keep the same iv Z translation 1 Need to compensate for 10 cm of travel during walk 2 Try not to translate the camera if we can avoid it 3 Try to compensate through tilt d Cameras i Resolution Mertz has 640x480 24 bit color ii Speed Mertz is 30 frames sec e Communication f
55. hment did not have weight information available and thus were approximated The specification of the lenses can be found in Appendix VII Analysis The motor torque was the key feature in selecting a motor The torque was calculated as the product of the angular acceleration and the mass moment of inertia Equation 1 T la 1 where T is the torque N m I is the mass moment of inertia kg m is the angular acceleration rad s The parameters for the motor calculations were taken from the specifications Table 4 By appropriately placing coordinate systems in the assembly of the head the mass moments of inertia were calculated by SolidWorks Table 5 Table 4 Parameters for motor calculations Parameters Min Ang Acceleration 1800 deg s Min Ang Acceleration 31 41 rad s Min Ang Speed 200 rom Min FS 2 29 Because the helmet was being designed in parallel the exact moment of inertia was unknown The helmet was assumed to be a thin sphere with a radius of 4 inches so as to enclose all the major pieces of the platform Equations 2 where I is the mass moment of inertia kg m m is the mass kg r is the radius of the sphere m This moment of inertia value was included as a rough estimate in the design calculations Table 5 Moments of inertia for Pan and Tilt Mass Moment of Inertia Tilt g mm Bum
56. ineers under their supervision to act in professional matter for each 50 employer or client as faithful agents or trustees and shall avoid conflicts of interest and to build their professional reputation on their merit of their services and shall not compete unfairly with others These three canons were not relevant to this design project The sixth canon states that engineers should only associate with reputable persons and organizations and this was also followed Finally as canon 7 states engineers should issue public documents truthfully and objectively This document satisfies both of those criteria The major source of controversy with this robot is that it would dehumanize the act of killing by having a remotely controlled robot commit the act This robot however is not intended to be used for warfare but for reconnaissance and intelligence gathering only In addition if used for its prescribed purposes the biped will save the lives of many men and women who serve in our armed forces In its current form and intended use as a demo for the IHMC biped the RVP has little risk of causing damage or injury This robot will be passed along to other engineers and they will be held responsible to similar if not the same code of ethics For our purposes we must trust our best judgment and the judgment of those in our military to appropriately use what we have developed Statement of Analysis of Safety Safety is a key part of the design p
57. ing controlled 49 ESC629ERAxisController This class is used to create two objects one represents the pan motor and another represents the tilt motor These objects contain simplified methods that often call methods from the driver class BucknellHeadOneController This is a class representative of the Bucknell head It creates the two ESC629ERAxisController objects and controls the desired positions based on feed from the gyro on the biped and or joystick The method doControl in this class is where adjustments would be made if a different type of positioning were desired BucknellHeadOneEmbeddedMain This is the class containing the main that is run on the PC 104 computer platform It sets up and creates the YoVariables for yaw pitch and roll A BucknellHeadOneController is also created in this class BucknellHeadOneSimulation This class sets up the Yobotics Simulation Construction Set GUI The GUI can be used to display and compare variables The joystick and slider board are set up in this class HeadinputDevices This is the class used in the BucknellHeadOneSimulation class that links either the slider board or the joystick to the head The code is written such that the slider board or joystick will only work if detected by the computer otherwise it will not affect the program Statement of Ethics There is a substantial amount of controversy surrounding defense related research particularly in the universit
58. ing to draw two feeds from the same camera instance PGR tech support suggested trying GraphEdit part of a the extras pack in the February 2005 DirectX SDK which allows users to manually choose audio and video devices and each step in the decoding and processing stream Each step is called a filter and hundreds of filter options are available depending on the hardware and software on the computer Help making a graph can be found at http www 3dtv at Products Multiplexer Installation_en aspxttRecording This and PGR support guided the production one of the more successful versions we used Figure 32 This set up also provided a way to simultaneously record and preview live video This did allow two separate PGR cameras to be set up through the Multiplexer and was successful in recording a side by side file that could later be played back in 3D in the goggles through the Player However the live preview window displayed in GraphEdit did not display in a suitable format for 3D in the goggles Additionally trying to open the live Stereo Multiplexer feed in the Player resulted in errors suggesting that cameras were not connected a format was not compatible or the signal was not being carried through all the filters Further efforts by PGR technicians were unable to resolve the problem and determined it was likely a problem with their own DirectShow filters or drivers and that they would work to resolve the issue in the future It remains unclear
59. ion Platform Initial Designs Based on existing robots Professor Shooter and the engineers at IHMC developed a list of requirements for the initial designs Both of the designs developed during the summer of 2009 were two degree of freedom systems They each used two brackets to move the cameras along the pan and tilt directions aligning the vision axes with the centers of rotation These designs 5 were fitted to mount Point Grey Firefly MV cameras However the motor selection and size for each design varied greatly The goal of these designs was to provide vision platforms that could be tested modified and further built upon Vision Platform Design 1 Kandler and Snyder The platform designed by Kandler and Snyder was designed to test the speed and acceleration requirements for the head It is a one camera system that utilizes two geared DC motors with optical encoders for position feedback The platform was designed to be as simple as possible The pan and tilt directions each had an aluminum bracket that attached to a motor directly by a screw clamp The motors and encoders were powered by and controlled with a PC 104 board that was able to integrate into the computer on the existing bipedal robot The controls were also integrated into the existing biped controls which allowed for active stabilization from the biped s onboard gyroscope One of the benefits of this aspect of the design is that the PC 104 could be used to power the motor
60. iously identified specifications Design Embodiment The final design for the simple rotation concept required some modifications to ensure the achievement of every specification Figure 14 There were two brushed DC motors geared down and attached to timing belts that rotate each axis The addition of a helmet was also included in the design Figure 15 E Figure 14 Final Design of RVP without helmet or wiring 22 As discussed in the Concept Design and Evaluation section above backlash and position accuracy were the extremely important Thus harmonic gearheads which were accurate to 1 30 of a degree were chosen in tandem with Micromotor DC motors There will be a hand machine piece that will mount the two pieces GT2 belts were employed to transmit the power and keep the backlash low These components are discussed in detail in the Component Selection Section of this report e Pan Bracket The pan bracket was determined to be a u shaped bracket composed of three individual plates This design was chosen for a couple of reasons A large single piece bracket would require a large block of aluminum and waste most of the material 23 Any large force on the arms could cause bending and lead to misalignment of the holes for the tilt shaft and require replacement of the bracket In addition assembly could prove difficult A three piece bracket allows for much easier assembly because the arms can be attached after other comp
61. irements the team developed a list of potential scenarios that the robot may encounter to give insight on defining some of the specifications List of Required Scenarios Navigating safely through an urban environment o Have a large enough range of vision to identify and avoid hanging objects steep ledges etc o Avoiding walking into walls windows and screen doors Going through a standard door 35 x 83 Going up and down stairs rise 6 9 5 Walking up and down sloped surfaces Gracefully walking over rough terrain Dynamically walking and balancing Recovering from a push Identifying and avoiding moving hazards something rolls or moves in its path like a dog or car Stepping over low obstacles door jams and curbs Interpreting and adapting to ground surface grass concrete carpet tile etc Seeing in color List of Potential Scenarios Navigating in low or no light Navigating in extremely bright light Navigating in low visibility such as fog or smoky rooms Seeing in 3 D Recording of visible history Obtaining audio feedback Operating in extremely high or low temperatures Operating in radioactive and or chemically or biologically hazardous environments Identifying traffic lights and other street signs Running jumping Ducking IHMC identified these requirements in the paper Design of a Bipedal Walking Robot Appendix 111 Additional Requirements from Design of a Bipedal Walking Robot Mec
62. king up into smaller groups can be more efficient Additionally the team has gained a lot of experience specifying motors researching transmission systems 3D vision and designing for manufacture The design required many hours analyzing various websites and catalogs to specify the appropriate motors and gearheads The calculations were fairly simple but actually finding a motor that fit the specifications was rather challenging and frustrating It was essential to realize that these kinds of decisions take time As we re always told but sometimes don t like to put into practice patience is a virtue Anything of quality takes time and effort Another aspect of the design process that gave us some troubles at first was conveying design concepts to each other without any 3 D models While trying to make decisions like how to mount the motors are design the pan shaft support we would discuss concepts and draw a lot of 2 D drawings on the lab whiteboards Often times this method led to confusion and misunderstanding When we were able to create actual models it was much easier to convey the information and make decisions Also it can be difficult to notice possible problems when just referencing a drawing Implementing the 3D stereo video proved to be a frustrating exercise This is largely because we were trying to push through using camera systems that were bought in the summer before the team was formed with no clear path on how to achieve stere
63. ld hit the PC 104 and RTD board when placed on the actual biped body Extra room along the sides should be sufficient to accommodate quick disconnect adapters for wires FireWire Hub Originally it was planned to place a hub on the tilt bracket to collect all the Firewires cables from the three cameras and produce one output wire This would minimize the number of wires going through the pan mechanism and make the wiring much simpler However difficulty arose because the hub limited the data rate through each connection to just 100 Mbps instead of 400 This was insufficient to run the Point Grey Firefly MV cameras any faster than 7 5 fps and was deemed unacceptable The specification was to have the cameras running at 30 fps The purchase of the Unibrain Fire i cameras eliminated the need for the Firewire hub since each camera contained 2 FireWire ports allowing the cameras to be daisy chained so just 1 cable to the computer was needed for the 2 cameras Camera Choices Setting up the vision system turned out to be one of the most difficult parts of the project The discovery of 3dtv s Stereoscopic Multiplexer and Player appeared to be the perfect solution for the project goals The system was designed to work with almost any webcam or digital video camera and was used successfully on the lab computer using a set of Logitech USB webcams and a set of Panasonic digital camcorders connected through FireWire With these 3D live video was easily r
64. llable Series Elastic Actuator These actuators provide high force fidelity and low impedance allowing for control techniques that exploit the natural dynamics of the robot The walking and balance recovery controllers will use the concepts of Capture Points and the Capture Region in order to decide where to step A Capture Point is a point on the ground in which a biped can step to in order to stop and the Capture Region is the locus of such points 1 INTRODUCTION To date there have been a number of three dimensional bipedal walking robots developed such as the Honda P2 P3 and ASIMO25 26 Sony QRIO27 Waseda University Wabian2s University of Munich s Johnnie29 30 Kawada AIST HRP 231 32 and others Many of these robots are limited to commanding joint positions rather than joint torques This makes it difficult to perform complex tasks such as walking blindly over rough terrain dynamically responding to unknown disturbances or possessing the ability to capitalize on passive dynamic control strategies Yobotics is developing a twelve degree of freedom bipedal walking robot platform which has been dubbed M2V2 see Figure 1 The robot will be capable of high fidelity force control at each of the 12 degrees of freedom With the ability to go beyond joint tracking trajectories M2V2 will be used to implement validate and extend various bipedal control algorithms including those developed on Spring Flamingo1 a planar biped and on a simulated thr
65. locks up 3D video works perfectly with 3dtv stereoscopic player and demo videos e Z800 goggles as mirrored monitor from Nvidia GeForce 6800 card o Use Nvidia 31 31 forceware and stereoscopic drivers Set 800x600 resolution 60hz o Play videos with 3dtv Stereoscopic player Output to Nvidia Stereo signal 76 Appendix VII Camera Lens Selection There are a few requirements for the type of lens that will be used on the robotic vision platform These requirements are compatibility issues with the Firefly MV cameras Firefly MV Specifications Image sensor type 1 3 progressive scan CMOS Mounting CS mount or C mount with adapter Max resolution 752x480 061 FPS Mass 378 The maximum image sensor size for the lenses must be at least as big as the 1 3 sensor which should not be a problem since many C mount lenses work with at least 2 3 The only difference between C mount and CS mount is that C mount has a 17 526mm flange focal distance and the CS mount has a 12 52 flange local distance This is the distance between the mounting flange and the firm place CS mounts typically a wider range of vision because of this shorter distance There are additionally some design specifications that are desireable and will serve as criteria for choosing the best camera lens Weight is one of the most important concern along with field of vision FOV and image resolution 3D Visor Specifications Resolution 800x
66. m throughout the year based on the previous head specifications and the successes and failures of each design Additionally each of the specifications had a corresponding validation technique for when the final prototype has been constructed The group modeled several different vision platform concepts in SolidWorks and evaluated their operation and performance as well as selecting the appropriate hardware for each A design was selected from this process and the first iteration of the RVP has been fully designed Fabrication of the RVP occurred in the second semester During the fabrication of the design modifications were made and are documented in the Discussion section of the report Testing procedures were created after assembly and these procedures were performed and can be found in Testing Procedures The team met weekly with the project advisors to discuss progress and determine goals for the coming week Meetings with the team members at IHMC are not as frequent but a couple teleconferences have occurred during the course of the project Requirements Beginning on the 20 of April 2009 and updated through the 29 of May 2009 a list of requirements for the head was compiled by Professor Steven Shooter Appendix Il Though these requirements were determined by a group of individuals from IHMC with significant 11 experience in both the robotic and design fields this list was vague in terms of numerical requirements From these requ
67. ming side of the project still leaves room for future work as well These issues include getting the gyroscope to function properly and consistently and working on enhancing the level of control It would possible and useful to include an algorithm that could account for translational movement using simple trigonometry and knowledge of depth of view Though it is far beyond the scope of this project there is also significant room for programming based on the camera signals It may be useful to develop object recognition and tracking that can be used to help the biped recognize something like a step or doorknob This would greatly enhance the capabilities of the biped Lessons Learned This project has been helpful to each of the team members on multiple levels The team has learned a lot about working in a team and communication Communication was critical amongst the team with the customer and with the project coordinator Communicating with IHMC was helpful because they were extremely clear about the components that were important to them and those that were unnecessary Because this is an ongoing project we had to ensure that our design was compatible with the equipment at IHMC Despite our infrequent meetings with them they were very clear about the specifications that were valuable to them It has been extremely difficult to manage time amongst the team which made communication difficult from time to time When working alone a schedule is much
68. n the bracket and clamp it around the shaft Electronic Implementation As with any design project modifications were made before fabrication Specifically the pan mechanism changed significantly from the previous design Additionally it was decided to add a hard stop as well as an optical sensor to protect the motors and gearheads should a bug in the programming occur Furthermore major changes in the base plate were made to accommodate the pan mechanism change as well as to provide space for three new circuit boards Electrical Connections The electronics involved in this Robotic Vision Platform were very complex The diagram below illustrates how the components were connected Figure 27 paneer Tilt DOF External Power Suppl 12v Motor Actuator 12v Ss Control Control Photogate Optical Encoder AAA gece Power amp Signal mA Signal Optical Encoder Photogate AS Bee Power e RTD Board Power E ear Signal to PC 104 Signal Bumblebee2 Camera Power amp Firefly MV Camera Firewire Control Power l TCP Ethernet ioe i Baseplate Gyro Firewire Hub Firewire Human Machine Interface a USB VGA Headphones Joystick 3D Goggles Figure 27 Electronic Connection Diagram Most of the electrical equipment ran through the PC 104 board and were then routed to the human machine interface The sensors on the tilt and the pan bracket as well as the base plate needed to be connected to the PC 104 to po
69. ne how fast the system actually moves using the encoders Validation Technique The final design will require some sort of encoders to give position feedback which can be used to measure the speed during operation Both maximum and average speeds will be measured Acceleration TBD Justification See above Validation Technique See above Accuracy of Positioning less than 1 deg Justification All of the zero backlash motors give this as a specification The IHMC head does not have a zero backlash gear head and it gives around 3 deg of backlash This is simply too much play and the motor tends to wiggle around A zero backlash gear head is necessary for the next design Validation Technique Any play in the motor is unacceptable Encoder Resolution 0 01 deg Justification Most encoders will be able to produce a resolution that is much better than this target value but this number serves as an upper limit This number is based on the Harvard KTH and TRISH heads Validation Technique This is a physical characteristic of whatever encoders we choose to use for the pan and tilt The value can be found on the encoder specification sheet Range of Motion 180 deg pan 150 deg tilt 72 Justification We want to have the head turn and at least look off both of the shoulders for the pan For the tilt the operator has no use looking straight down at the body but may want to look straight up Therefore 90 deg up and 60
70. o video 3D goggles and software were later chosen under the assumption that the PGR cameras were sufficient without checking for format compatibility another concept that was not fully understood Simple webcams proved to be more compatible than the expensive cameras a realization that took months to come to which was much needed time that could have benefitted other parts of the project If starting over it would be essential to check to make sure that all electronic components are compatible with each other before investing time and money in them 63 Allocation of Work Scott Bevan Initial and final shell design part drawings selected belts microphones carbon fiber speakers 3D goggles fabrication shell fabrication Matthew Kandler Assembly designed shafts specified pulleys camera lenses and bearings fabrication programming Danielle Renzi motor calculations specified gearheads thermal analysis sustainability analysis bill of materials fabrication design amp fabrication of testing apparatus William Rittase Drawing files specified motor motor assemblies exploded view fabrication shell design shell fabrication emotional support 64 Appendices Appendix I Stereo Vision Camera and Goggle Research Stereo Vision Cameras Brand Product Description Cost Bumblebee 2 0 BB2 0352 IEEE 1394 FireWire Interface Sony 1 3 progressive scan CCD 12cm baseline 648x488 at 48fps
71. object is created in the embedded main that controls and represents the entire robot head setup See code for more detailed information and documentation 47 Compiling and Running the Software See User s Manual for in depth instructions Explanation of Controls Currently there are three control modes set up for the Robotic Vision Platform All three modes are accessible through a single jar file EmbeddedMainGyro jar Modes are selected by adjusting the MODE dial on the right side of the joystick The first mode uses input from a MEMSense gyroscope mounted on the back of the base plate of the RVP Rotational data from the Z and Y axes of the gyro are used to stabilize the RVP when rotated This is done simply by setting the desired positions to the opposite of the rotational position i e if the gyro reads 0 5 radians in the Z axis the pan will be set to 0 5 radians In addition control from the joystick is overlaid on top of the gyroscope control This allows for the operator to control the position of the cameras while maintaining a stable image The second control mode works the same as primary control mode but without the gyro control and more sophisticated joystick control In this mode the joystick is by default set to control the velocity of the head So if the joystick were to move to the left the head would rotate counter clockwise for as long as the joystick was held in that direction By holding down the FIRE button on
72. onents If either arm were to bend and require replacement it could easily be replaced Also if the team decided to change the tilt bracket if the camera layout were changed for example the arms could be switched out This design enhances the ability to improve and iterate on the head Tilt Bracket The tilt bracket was constructed to fit a BumbleBee2 and two Firefly MV cameras To allow for adjustability the Fireflies were mounted to slots The ideal distance between the cameras is not yet known and needs to be tested by using the slots to adjust the cameras to optimize the distance for 3D vision Additionally the dimensions of the tilt bracket were refined to minimize the moment of inertia The length of the sides was adjusted to set the center of gravity about the shaft holes Motor Slots Proper tensioning of the belt drives was important In order to achieve tensioning slots are included for the motor mounts Sliding the motors will provide an easy way to tension while maintaining alignment Tilt Shaft The tilt axis was divided into two sides the drive side and the supporting side The supporting side of the tilt axis was constrained radially using a shoulder bolt and flanged ball bearings Shoulder bolts were precision machined to improve alignment A stainless steel shaft was to be press fit into the other side of the tilt bracket This however was changed during fabrication and is addressed in the Discussion section This sha
73. ositioning Resolution of 01 004 Positioning Range of Motion 180 pan 175 pan 150 tilt sufficient tilt Number of Cameras 4 4 Camera Frame Rate 30 frames per second 30 Field of Vision 60 39 0 Resolution 640 x 480 640 x 480 Not all of the original specifications were met The size is roughly within the initial constraint but does exceed its specified width Additionally the head weighs roughly 8 Ib more than was intended The design was on course to achieve 4 Ibs or below but electronic equipment changes to the motor gearhead piece changes to the base and other issues added unexpected weight Though this is significant there are ways that the weight can be reduced which are discussed in the Future Work section The speed and acceleration were testing for each of the axes The pan axis showed slightly slower speeds and accelerations Figure 39 The tilt axis moved a little bit faster Figure 40 This is likely due to the additional gear reduction introduced in the pan by the pulleys 58 200 E gt c a Pan Accel 150 s pea Apoan 10 200 15 Time seconds Figure 39 Measured Velocity and Accerleration for the Pan Axis 59 150 Tilt Vel Tilt Accel 100 50 gt 2 E o Kach e gt 0 Q 2 K o 2 y gt E KI 0 10 100 15 150 Time seconds Figure 40 Measured Velocity and Accelerations for Tilt Axis The speed originally s
74. outed though the Multiplexer to the Player and goggles Despite the simple 40 straightforward configuration used with these cameras much of the semester was focused on getting the Firefly MV cameras to work with the Stereoscopic Multiplexer and Player Initial testing of the Firefly MV cameras simply involved making sure they worked in various combinations within the supplied Point Grey Research PGR Flycapture software Not until the FireWire hub and special 6 to 9 pin 1394 cables arrived was the bandwidth problem discovered with the hub After that it was attempted to view the cameras in other programs such as the Multiplexer and Player However the default PGR Software Flycapture and drivers tended to lock the cameras to PGR software thus making them invisible as a camera to the computer and other programs Instead of showing up as imaging devices in the Device Manager they were Point Grey Research Devices After working with PGR this limitation was overcome by installing a new 2 0 3 version of the driver and making some registry changes documented in the file Registry Change with PGR Flycapture 2 This allowed the computer and the Multiplexer to see PGR cameras but only indicated that one at a time was present If just one was connected the Multiplexer would begin configuration and then fail because it wanted a second camera lf two PGR cameras were connected configuration would not progress presumable because it was try
75. pecified was 200 rpm The maximum speed obtained was only 95rpm This was most likely due to the limited range of motion As can be seen the acceleration was significantly higher than specified reaching 6900 deg s It appeared that the RVP could attain higher speeds if it had more time to reach them Additionally 95 rpm appeared to be sufficiently fast and if it were to go faster it would likely cause the operator discomfort This high acceleration is expected to be able to reduce the shake in the camera view to only high order vibrations These can be completely eliminated by introducing image stabilization software Finally the field of vision was less than originally expected The field of vision was only found to be about 39 A picture was taken of a white board at a fixed distance away Figure 41 The width of the actual white board was compared to the distance away from the cameras to determine the field of vision using simple trigonometric relationships Table 14 The 60 figure was just an estimate based on the cameras which were used over the summer but it appears as though the 40 field of view was sufficient for the operator to get a good view of his surroundings without much distortion of the image If it were desired to have a greater range of vision the camera lenses could be easily replaced for ones with a greater field of view This was not done because of the significant distortion introduced by increasing field of vie
76. pports the Nvidia output This software and the related Multiplexer software from 3dtv was focused on as the methodused to capture transform and output the video signals so that they can be displayed by the goggles in real time 28 Statement of Analysis of Motor Specification Problem Statement The task of finding a motor proved to be very difficult The motor needed to be able to move the vision platform quickly while not adding much weight to the design The motor also needed to be reversible Preferably the motor was to be connected to an encoder and would not be backdriveable Finding a motor and gearhead combination that achieved the desired torque and speed with minimal backlash was especially difficult for the range of size and weight The specifications indicate that the entire head needs to weigh less than 4 pounds and fit inside an 8 x8 x10 box Additionally the head needed to be able to move with an angular acceleration of 1800 deg s in both directions Assumptions A variety of assumptions were made in the calculations to analyze the motor The exact weights of the final components were estimated because the design was being refined during the motor selection process Also the helmet that will be used to protect the head was not completely designed thus it was approximated as a thin sphere with a radius of 4 inches and a weight of one pound The screws shoulder bolts camera lenses hub wires and other forms of attac
77. r was specified to achieve 16 mNm at normal operating conditions Additionally an optical encoder was placed directly on the back To increase the torque a harmonic gearhead was chosen This gearhead was chosen because it almost completely eliminates backlash while increasing the torque With a gear ratio of 50 this combination is capable of running at 800 mNm This motor and gearhead combination provided a factor of safety of 3 67 for the pan mechanism and 7 35 for the tilt mechanism Both of these values were well above the specified factor of safety of at least 2 The motors could achieve the torque but they also needed to be able to move quickly The maximum angular acceleration of the motor was specified at 140 000 rad s With the gear ratio specified this is about 160 000 deg s Additionally a minimum speed of 200 rpm was required This specification however was not met The no load speed of the motor was 8100 rpm but with the gear ratio this is reduced to only 162 rpm Initially this was a concern but after building the platform the speed of the head is sufficient In fact a speed much faster than the current motion would likely cause discomfort to the operator looking through the cameras Tiny motors were necessary because of size and weight constraints The larger concern was that the large gear ratio would completely negate the speed and acceleration of the motor The reason that the Micromotor was chosen was because it would be able
78. rocess It is extremely important to design for safety when designing a piece of equipment that will be used as a demonstration but also involving expensive equipment The safety of the people interacting with the robot and the protection of the equipment were considered in this analysis To protect bystanders there is a helmet built around the equipment If something were to break and eject off of the body of the vision platform the shell would block this piece from hitting anyone Also the helmet protects people from putting their fingers inside and getting them caught on moving parts The team members also designed to protect the equipment A minimum factor of safety of 2 was used in the motor calculations to compensate for any friction in the system inaccurate assumptions and other inaccuracies in calculations When the final calculations were done and the motor was specified the factor of safety was higher than 2 The final design calls for a gearhead with a 50 1 ratio which yields a factor of safety of 7 35 for the tilt torque and 3 67 for the pan torque Additionally the moment of inertia calculations estimate a shell weighing a generous 1 pound The helmet is designed to be as light as possible and is also made from carbon fiber so it will likely weigh much less than this Additionally one extra bearing of each size is being ordered as well as a range of sizes of belts to ensure that the vision platform will work exactly as it should Fur
79. s of backlash to occur the machined mounting positions would have to be extremely precise Any small misalignment could completely negate the effects of the low backlash gears Cable Drives A more robust transmission option was the cable drive Depending on the material of the cable the drive could handle extremely high torques up to 76 000 Ib ft Since the cables are a flexible material the position of the drive has much more freedom than a traditional gear set In addition to their flexibility cable drives have extremely low backlash as long as they are properly tensioned they do not require as precise alignment because of the tensioning in the cables Though cable drives minimize backlash they are not able to provide significant gear reduction without sacrificing size This could be rectified by using a gearhead on the motor Another concern with the cable drives is that the tension on the pulleys needs to be accounted for when designing the shafts for the drives Belt Drives Timing belt drives were also considered as a potential transmission system Timing belts like cable drives minimize backlash without requiring high machine tolerances Belt drives are used in many rapid motion applications such as printers and laser cutting machines A benefit of belt drives is that the belts and pulleys can easily be purchased in standard sizes One issue is that the head assembly must be considered while designing in belt drives Belts cannot just be
80. s of future designs With minor adjustments in the code to account for new encoder resolutions and gearing the same program could be used to run the next design This head was eventually mounted on the biped in the IHMC lab Figure 3 After running several tests on the active stabilization it was found that the speed and acceleration were more than adequate to stabilize the camera during motion One of the primary concerns with this design was backlash in the motors Though the testing proved that the stabilization was sufficient there was still some backlash in the gearheads which were mounted on the motor by the supplier According to the specifications of the gearheads there was up to 3 of backlash This meant that position of the camera could never be accurate to within 3 Figure 3 Kandler Snyder RVP Vision Platform Design 2 Rittase and Sirot The platform developed by Rittase and Sirot used RC servo motors as the primary actuation instead of the DC brushed motors used in the Kandler Snyder design Figure 4 These motors were given commands to go to an exact location by pulse width modulation and therefore did not require any positioning feedback device such as an encoder A joystick was used to control the position of the camera system Figure 4 Rittase Sirot RVP Two support brackets allowed for the head to move in both a pan and tilt motion In order to accurately balance the moment of inertia the camera mount had slo
81. sues with the joystick in the Rittase Sirot design and issues with the PC 104 board in the Kandler Snyder design Because the team was familiar the basic benefits and drawbacks of each design it was determined that time was better spent designing the next iteration instead of testing the old designs Specification Testing Testing procedures were developed for each of the specifications that were identified at the beginning of the project 8 months ago Table 12 Additionally it is imperative that the head can compensate for the motion of walking Testing procedures were also developed to examine 55 this feature The RVP was first tested without a helmet and then again after the shell had been manufactured Table 12 List of Specifications Specifications Value Size smaller than 8 x8 x 10 DxWxH Weight 4 pounds DOF 2 Min Speed 200 rpm Min Acceleration 1800 deg s Accuracy of Less than 1 Positioning Resolution of 01 Positioning Range of Motion 180 pan 150 tilt Number of Cameras 4 Camera Frame Rate 30 frames per second Field of Vision 60 Resolution 640 x 480 Size Use ruler Weight Place on scale Speed The RVP was set at an acceleration of 100 rad s and moved back and forth within its range of motion Encoder data was used with the Yobotics software in order to record position with respect to time Speeds were determined using the known gearin
82. summer of 2009 one of which is a single integrated camera system while the other is the combination of 7 multiple discrete cameras Appendix I Both of these solutions were purchased from Point Grey Research PGR before the start of the semester The team worked through April under the assumption that these cameras would be implemented in the design It was later found however that due to inconsistencies with Point Grey software and the image format they output it would be impossible to use two Firefly MVs to obtain a 3D image with the Stereoscopic Multiplexer and Player In place of the Firefly MVs two Unibrain Fire i cameras were purchased The three camera systems are briefly mentioned below and the decision between them will be discussed in detail in the Electrical Implementation section The first system is called the Bumblebee2 and is an integrated system that contains two cameras mounted in a single case that is calibrated for 3D vision and mapping Figure 5 It is a comprehensive solution that includes software for camera control image rectification 3D point mapping object tracking and multi camera integration The sufficiently high resolution and frame rate meet the engineering specifications for the vision system Its main drawbacks are its relatively high weight 0 75 Ib large size 6 2 in long and unknown capability to create a 3D video output signal In addition to the Bumblebee two Firefly MV cameras were included in
83. t g 111 210 139 116 23 Input diameter mm 3 3 Max Torque 2500 6000 4500 1000 Cost 596 00 273 40 215 80 99 50 Resulting Speed rpm 283 172 174 174 125 Torque mN m 480 464 368 368 640 FS Pan 2 2 2 1 1 7 1 7 2 9 FS Tilt 4 4 4 3 3 4 3 4 5 9 79 Notes Moog Gearhead This gearhead is roughly the same size as the currently speced gearhead Harmonic Drives LLC It is 61 mm long where the harmonic drives on is 51 mm long The shaft inputs are both 3 mm The faces of harmonic is 32mm x 32 mm and this isa 32 mm diameter weighs less but significantly more backlash MicroMo The S signifies that it must be custom made from steel 80
84. tant for the vision platform Any play in the motors will cause unstable images and inaccurate positioning This error will propagate through the stereo vision algorithms creating error in the position of the physical objects In addition the image platform will shake while walking causing discomfort for the operator Because of this the accuracy of positioning was specified to be below 1 degree Minimizing weight is another important design consideration IHMC has asked that the total weight of the head be kept to no more than about 4 lbs to reduce the load on the body The Bumblebee2 and two Firefly MVs that have to be attached to the head have a combined weight of 832 lbs leaving about 3 lbs for the metal framework motors shell and Firefly camera lenses Keeping the actuated mass at a minimum will reduce the load on the actuators thus allowing for smaller motors and an overall lower weight This will also help to improve the speed and accuracy of head motion In addition to maintaining a low weight the vision platform should also be small and humanlike in shape A human head is approximately 7 4 high and 5 wide The specification determined for this platform is to fit within a box 8 x 8 x 10 This is much bigger than a human head to account for the wide shape of the Bumblebee2 and also to include room for the electronic equipment Additional information on the justification and plans for validation of the specifications are
85. the joystick the joystick control switches to position control When the joystick is in the stable position the head is looking straight forward and the position of the head will adjust with the position of the joystick A convenient feature of this is that if the head is desired to be returned to center the user can simply hold the joystick in the stable position and hold the primary trigger In this mode the gyro is still calibrated and the signals are displayed in the Yobotics GUI but they are not used for any control In the first two modes the joystick throttle can be used to adjust the velocity and acceleration of the motor control Lastly the third mode is set to control the RVP based on the position of the mouse on the screen The X direction moves the head in pan and the Y direction moves the head in tilt This mode is intended to be a step toward attaining control from the gyro in the goggles This gyro is connected to the mouse input and thus it would be able to control the head in this mode An operator would not have to worry about using one joystick for the biped and one for the head Spring Semester Programming Issues Most of the control for the head was carried over from the work performed by Kandler and Snyder while working at IHMC However some changes and additions had to be made to allow for additional features and changes in hardware The first change was to adjust the encoder resolutions and gear ratios for the two axes This
86. the radius from the tilt axis to the bottom of the pan bracket thus making it unnecessarily bulky The clear shield would also be very difficult to keep optically clear and could bend the image Figure 18 25 Figure 17 Anthropomorphic Shell Figure 18 Spherical Shell A third design took both of these into consideration Both of the camera systems would have a hole directly in front of the lens to give the appropriate vision There was an inside covering over the cameras which tilts independent from the outside housing which is connected to the pan The issues with this were in the separation between the tilt pan housings There is room for fingers or other objects to get caught when in operation In addition the lenses are exposed to the environment since they are the most important parts to the vision aspect of the project they must be protected It also looked rather odd as far as aesthetics were concerned Figure 19 Figure 19 Semester 1 Final Design The final design was much simpler than the previous design The result was a fixture which moved only panned This removed of the possibility for the tilt to interfere with the pan In addition it added to simplicity of removal of the shell from the head to be able to easily access 26 the hardware underneath the shell In addition the ease of manufacturability and assembly of the shell was integral in this design There was only one front and one back which attached to the pan
87. thermore calculations were performed to ensure that the belts would not break by the loads applied by the motor and gearhead The stall torque was used in calculating the maximum torque that the motor could supply to account for the worst case scenario The stall torque is 51 rated to be 08 N m and when the gearhead is considered a 4 N m torque could potentially be inflicted on the pulleys Table 7 Table 7 Maximum Torque Calculation Stall Torque 0 08 N m Gear Ratio 50 Max Torque 4 N m The force that would be felt by the belt was determined by the maximum amount of torque the motor could supply 4 N m and the radius of the pulley Table 8 Table 8 Potential Force Applied on Belt Diameter of Pulley in Diameter of Pulley mm Force Felt by Belt N 1 003 25 314 1 2 30 262 The breaking strength of the pulley was rated at 86 N per mm of width It was determined that a 3 mm belt would not be able to handle the forces that the motor could apply for either size pulley Table 9 A 6 mm pulley has a factor of safety of almost 2 for a 1 2 diameter pulley and 1 6 for a 1 003 diameter pulley Considering that the motor should never be running at its stall torque a 6 mm belt width with a 1 003 diameter pulley should be more than satisfactory for this application Table 9 Belt Breaking Strength Analysis 2 mm Pitch Actual Force on 1 003 in Ac
88. to achieve the necessary torque and still keep the acceleration high enough Statement of Significance The calculation of the torque was of great significance to this project This calculation required the collaboration among almost all of the specifications listed Once the required torque was calculated it was a matter of finding appropriate motors that were also compatible with zero backlash gearheaos as well as encoders Component Selection Two major components were selected in the design of the RVP The combination of the motor encoder and gearhead was essential to satisfy the speed acceleration and accuracy of positioning constraints Additionally the timing belt selection was also vital in achieving these specifications The motors were Series 2342 S0 12 CR from MicroMo with optical encoders OPECO7 attached on the back These encoders are indexing encoders that determine the position of the motor 31 relative to an initial home There was one low backlash gearhead attached to the shaft of each motor These were harmonic gearheads Series CSF 08 50 2XH F from Harmonic Drives LLC The gearheads had 2 arc min of backlash or the equivalent of 1 30 of a degree which was well below the previously specified 1 degree The gearhead had a 50 1 gear ratio that was used to achieve the appropriate torque The calculations and requirements for torque were discussed in depth in the Statement of Analysis of Motor Specification The motor output s
89. ts to manually balance the system by varying the position of the two cameras both Firefly MVs This helped in the motor selection The team could accurately position the center of mass at the axis of rotation to decrease the torque requirements on the motors This also put less axial load on the motors once the power was turned off This system performed adequately but was also hindered by the backlash experienced in the pan and the weak tilt motor Since the plastic shell casing was not included in motor sizing calculations removed in the figure above the specified tilt motor was not able to provide enough torque The motion of this sytem was not smooth enough due to the discrete nature of servo motors and even when more powerful motors with metal gears were used over their plastic geared counterparts the backlash was still slightly too high to meet project requirements The motion transmission system differed from the Kandler Snyder design which had its motors directly connected to the axis of rotation This design had a push rod system to move the tilt actuation away from the center of rotation Therefore there was not a motor hanging off the side which may have gotten in the way of a shell or helmet In addition the motor was moved closer to the pan axis of rotation decreasing the moment which the pan motor was required to accelerate Three Dimensional Vision Two methods of creating stereo vision emerged from research performed over the
90. tual Force on 1 2 in Pulley N Pulley N 314 262 Fact f Safety for 1 003 Fact f Safety for 1 2 Width mm Breaking Force N SE E Ar a E in Pulley in Pulley 258 0 82 0 98 516 1 64 1 97 774 2 46 2 95 This design includes expensive and fragile equipment A number of measures were taken to ensure that this head would be run sustainably and would protect the components Calculations were performed to ensure that the motor would not overheat The terminal resistance of the motor and the torque constant were used to calculate the current Equation 3 52 I 3 where I is the current Amps M is the loading torque mN m Ky is the torque constant mN m amp From the current through the motor and the resistances in the motor the temperature increase can be obtained Equation 4 Tine I R run Rad 4 where Tinc is the increase in temperature across the motor K R is the terminal resistance of the motor Q Rau is the thermal resistance from windings to case K W Rryz is the thermal resistance from case to ambient K W The motor parameters were provided by MicroMo Table 10 Table 10 Motor thermal parameters Variable Value Unit Torque Constant KM 26 1 mNm amp Toque at Winding Mo 436 mNm Thermal Resistance thru motor Th1 3 C Watt Thermal Resistance case to ambient Th2 15 C Watt Terminal Resistance R 7 1 Ohms From these paramet
91. tures On Off Brightness Contrast Focus and IPD AC Adaptor Included 110 130V AC or 220 240V 3995 CyberMind Visette45 e Dual Input SXGA 1280x1024 12 900 SXGA e 45 deg FOV e Can pretty much customize to whatever you want Kaiser ProView SR80 e Dual Input 27 500 e 80deg FOV e Fits 5 of females 95 of males e Ifyou care to know more see the price Multiplexer 3D Encoder Name Description Price Dimension Technologies Inc Two video inputs s video output 2500 3D Video Encoder Either field sequential or side by side 67 Appendix II Vision Platform Requirements Objectives 1 Physical Head Characteristics humanlike head a Height humanlike about 18 7 cm 7 4 inches b Width humanlike about 15 1 cm 5 inches c Eye distance humanlike about 6 1 cm 2 4 inches d Weight less than 4 25 lbs Mertz head weighs that minimizing weight reduces load on actuators and can increase reliability Also reduces weight on biped e Degrees of freedom 2 turn and bob pan and tilt robonaut has side tilt but found it unnecessary and don t use it If head moves do eyes have to f Vibration isolation g Aesthetics i Helmet humanlike or other ii Characteristics that make face have more humanness are eyelids nose and mouth source diSalvo Note these aren t functional for our application 2 Vision characteristics a Binocular A lot of t
92. uble Channel Stereo Power Supply 1 000 mAh Rechargeable Battery Life Approximately 3 5 hours Weight 2 4 ounces 379 95 i glasses i3TV Resolution 800 x 600 1 44 Million Pixels per Display Field of View 26 Degrees Diagonal Virtual Image Size 70 at 13 Color Depth 256 Levels per Color True 24 Bit Contrast Ratio 75 to 1 Focus 13 TBR Eye Relief 25mm Exit Pupil 17mmH x 6mmV Convergence 7 10 100 Overlap TBR Refresh Rate Flicker Free 100hz display rate Audio Full Stereo PAL NTSC SECAM Composite or S video Input Input Frequency 50 or 60 Hz 25 or 30 Hz Interlaced 899 95 eMagin Z800 w Resolution 800 x 600 Pixels 1 44 Million Pixels per OLED Display Colors 16 7 Milllion Field of View 40 Degrees Diagonal Virtual Image Size 105 at 12 Tracking Built in 3 axis head tracking 1 deg accuracy Inputs VGA frame sequential 3D video 60 Hz dual input version available Compatibility Win XP most ATI and Nvidia cards Extras Built in headphones and noise cancelling microphone custom colors Weight 8 oz headset 1499 Sale 1299 66 Visette Pro Resolution 640 x 480 Pixels 920 000 Pixels per Display Field of View 60 Degrees Diagonal Eye Distance 60 70 mm adjustable Stereo 2 independent channels no sync needed Inputs VGA Composite NTSC or PAL Weight Approx 840 g incl battery Adjusts to Fit all Individuals Control Fea
93. verify the critical specifications of the previously designed vision platforms are defined below The values obtained in this testing will be compared with the vision platform that is currently being designed Specification Torque 1 Attach a force gauge or spring scale at points noted in Figure 37 and Figure 38 2 Run motor until stall 3 Repeat for several motor voltages DC motors only 4 Torque is calculated from Fxr Figure 37 IHMC head design testing locations 54 D Figure 38 Bucknell head design testing locations Specification Speed Acceleration Move head at max speed Do pan and tilt both separately and simultaneously Record encoder data on IHMC head during movement to find position velocity acceleration Use video camera to record movements of IHMC and Bucknell heads Use MaxTraq to find speed and acceleration of Bucknell head and verify encoder data on IHMC head Specification Power 1 SH Pw Pa Remove camera from IHMC head Attach spool and string to shaft of horizontally mounted motor Attach mass to other end of string hang off table Measure time required to pull mass up set distance P Fd t Repeat for several motor voltages Specification Durability 1 After completion of other testing set up both heads to run back and forth for a set period of time Periodically monitor heads to inspect for damage wear or other problems with components These plans were not acted upon because of is
94. w 60 PGR FlyCap Firefly MV FFMV O3MTC 9110490 File View Help DA Processed FPS 30 01 Hz Image 640x480 Cursor n a RGB n a Figure 41 Field of Vision Testing Screenshot Table 14 Field of Vision Distance from Lens 24in Width 15 75 in Horizontal angle 39 0 System Results The overall system test could not be performed because of problems with the gyro described in the programming section It is likely that full system testing will not be able to be performed until the head is delivered to IHMC and testing can be done while mounted on the biped The issues with the gyro should not be present in that setup However the results from the testing thus far supports that there will be positive results from complete system testing As stated the accelerations are more than sufficient and should be able to compensate for any major movements Future Work In the future it would be ideal to attain the values listed in the beginning of the project Primarily the size and weight of the RVP need to be lessened A number of solutions could be implemented to reduce the weight Pockets could be drilled into the pan base on the opposite side of the motor Additionally the motor and gearhead connection pieces are fairly hefty These pieces were made thicker because of unexpected dimensions of the gearhead If the gearheads had been in the lab before the gearhead and motor connection piece was designed
95. was accomplished by adjusting variables in the Axis Controller class The controls as previously mentioned have become somewhat more sophisticated than the programming from the past summer Originally the head could only be controlled by the 48 joystick in position mode or with the existing gyro on the biped More of the joystick inputs were explored to enhance the controls to the current capabilities Most of the programming this semester involved receiving and interpreting the data from the gyroscope Several classes were imported from the YoboticsBiped project to take the raw signals and process those signals to get rotational data These classes need to be cleaned up and gutted to remove any extra lines such as those controlling the various degrees of freedom on the biped Only the gyro data was desired to be taken from these files The gyroscope has been giving a variety of problems In general the gyroscope can perform correctly but there are a couple issues that arise with the Z axis during control The first issue occurs whenever the system receives a significant shock or vibration Any shock causes the Z axis to make a slight jump depending on the magnitude of the shock This clearly was not an issue with the head designed at IHMC since the biped experienced significant vibration during testing The other problem is that the Z axis begins to drift when rotating quickly in a back and forth manner This creates an offset whic
96. wer the sensor as well as send a signal to the computer The motors were controlled via the PC 104 and received power externally The PC 104 was connected to the Human Machine Interface via an Ethernet cable The cameras will be connected to a FireWire hub to simplify the number of cables that will be twisting around the head but these will run back to a Windows PC for the visual Human Machine Interface Currently 2 separate Windows PC s are used for the video and Yobotics joystick control 36 although the team will attempt to consolidate this to one computer for the Expo and future operations A wiring diagram can be found in Appendix IX Power from AC Adapter to PC 104 Control and Power from PC 104 to Pan amp Tilt motors Signals from Pan amp Tilt encoders to PC 104 Figure 28 Connections to PC 104 Pan Mechanism The main reason for the change was to accommodate for the wires coming from the tilt motor cameras and sensors above the pan base There was a major concern that the wires might twist around the pan base the pan belt parts of the old pan mechanism or even around themselves and thus snap at their connections The new design has a inch hole through the axis of rotation the wires will be fed through this hole and down through the base plate to the PC 104 The chance of twisting around parts and possibly breaking at their connections has been significantly lessened by the new design Figure 29 3
97. xes there is also the potential to perform balancing such that the rotation axes are collinear with the center of gravity This would greatly reduce the moment of inertia and decrease energy requirements One of the issues with this concept is that the tilt motor must be attached to the pan axis Thus the pan motor must be able to drive the weight of the tilt motor and the wires will not be stationary coming from the tilt motor Also the cameras will not be able to see the feet of the biped By looking directly down and not moving forward the cameras would be looking into the body of the biped rather than at the feet Neck Joint The neck joint is a more anthropomorphic approach to the conception of the robot head There are a few major benefits of this design The neck lifts the head above the axis of rotation which would allow the head to move forward while looking downward and possibly be able to view the feet of the biped There are two potential concepts of the neck design One of these designs Figure 12 allows for parallel motion of the head which means that the pan and tilt motors could both be mounted on the stationary part of the head Since neither motor would be moving the other the motors could be sized down This would make the wiring of the motors much simpler as well The camera platform would also allow for easy attachment of a shell or helmet The platform has plenty of mounting space and does not interfere with other parts
98. y high gears have p Heavy Gears backlash of machine pretty E much anything Unlikely slippage 1 3 arc min Noisy at high speeds bulky After analyzing backlash of the various types of transmission it was noted that backlash was not an issue for any of the transmission systems Each of the options could have very low backlash if properly designed The big gears and worm gear options however would require much more precise tolerances and assembly to reduce backlash 18 The belt and cable drives would be the most ideal choices in terms of weight Belts and cables would have comparably negligible weight and pulleys would not be nearly as heavy as the gears used in the other options Belts and pulleys also make the layout of the design much simpler because they can be mounted further from the axes of rotation without adding significant weight and thus altering the center of gravity Cable drives are very useful for specific applications but are complicated to machine and design For these reasons listed the belt drive was determined to be the preferred method of transmission The main concern however will be to find a way to properly tension the belt Concepts The next step was to start exploring implementation of the transmission systems Though there are an infinite number of solutions to the design problem in order to maintain reasonable specifications and performance two primary concepts were seriously
99. y setting This team feels however it has acted in an ethical manner and that this project is ethical The team has followed each of the seven ASME Fundamental Canons The first fundamental canon provided by ASME dictates that the robot should ensure the safety of the public The purpose of the fully developed robot is to operate in an urban society and provide reconnaissance remotely This robot is not designed to kill destroy or hurt human beings in any way in accordance with ASME s 1 fundamental canon Additionally in accordance with the second canon we have only done engineering work within our area of expertise When the team did not feel comfortable designing a component the team consulted specialists Primarily Jeff Gum was of great assistance in matters concerning electrical components Also Jason Geist Brent Noll Dan Johnson and Matt Tanner greatly assisted in the fabrication and design of the mechanical components In terms of analyzing the programming John Carff and Jerry Pratt were consulted Component manufactures were also contacted to help troubleshoot problems and suggest set ups Finally our advisor Prof Shooter was also consulted for safety and design calculations and general questions throughout the design process The third through fifth canons are as follows to continue their engineers professional development throughout their careers and shall provide opportunities for the professional development of those eng
100. y were tested on a lab computer The lab computers do not have the recommended Nvidia video cards that are natively compatible with the goggles See Appendix VI for a list of all the relevant computer components and options After consulting with the goggle retailers it was decided to try using them with the current ATI card and special drivers from monitor maker Z3D After trying many settings a demo screen in the driver had the appearance of trying to display in 3D but there was noticeable flickering and horizontal scan lines that detracted from the effect A short time later the computer crashed and stopped recognizing the goggles this problem could not be resolved in the lab Over Thanksgiving break a team member tested the goggles on his home computer using a recommended Nvidia GeForce 6800 card Using 91 31 forceware and stereo drivers allowed easy use of 3D mode The driver demo worked flawlessly and 2 games worked with varying 3D effect The stereoscopic player from 3dtv was also installed and used to watch numerous demo movies with good effect and no problems Because of the problems experienced trying to get the ATI video cards in the lab to work with the goggles and the demonstrated success of an Nvidia card the team procured two Nvidia 6800 video cards to install and use This simplified the process greatly and allowed the team to consult the goggle manufacturers for support Additionally the 3dtv Stereoscopic player also natively su

Download Pdf Manuals

Related Search

Related Contents

LED LCD TV - s3.amazonaws.com  Station météo avec prévisions, capteur de  情報アクセシビリティの国際標準化に関する調査研究(PDFﾌｧｲﾙ約  L`ÉDUCATION A L`ENVIRONNEMENT EN ITALIE dans le cadre des  www.philips.com/welcome  組み立て説明書（PDF）  Rapoo 6080  Manual de Instrucciones  ACADEMIA MILITAR  

Copyright © All rights reserved. Failed to retrieve file