Goal User Manual - Interactive Intelligence group
Contents
1. 4 6 Distributed Running of 4 6 1 Messaging Infrastructures 15 2 Distributing MAS 22d ke a eU Debugging a Multi Agent System D L Introspecting Agents 1 2 Rer UR SR TORY e ER a mE Y 5 2 Querying a Mental State 2 222r 5 3 Executing an Action using the Introspector 5 4 Debugging Output oa c a ek ee 9 91x Roe PO RR Pre Pe 45 a A Action Log onde ke Re Ede cn s Debug Trier 2 0 noe eq ode Roe he mom no Eo Customizing tha 222222622333 9 5 5 petting Breakpomte 222223493 39 X he E Y 5 5 1 Stepping and Global 5 5 2 Setting Code Line Breakpoints 0000 l 11 11 12 12 13 14 14 15 15 16 17 17 17 18 18 18 18 CONTENTS 5 6 Prolog Level Debugging so rarei na RR Romo mom thm m Re 29 5 6 1 Exporting a Mental 29 6 6 2 Prolog Exceptions 222224999999 x xo RE YR eS 30 5 7 Runtime Preference 31 Si Platform Management 2 46 00 ro r ROME EEST T TTE 31 mf 205093 99 x b e e yo e oe e dx b e Ox M 31 Running in Stand Alone Mode 33 6 1 Setting the JVM2. Chapter 3 Creating and Managing Multi Agent Projects This chapter explains how to create new and manage existing multi agent projects 3 1 Creating a New Multi Agent Project How to create a new multi agent project e In the GOAL perspective in Eclipse select File gt New GOAL Project e Enter a name and optionally browse for an environment file to include in the project e Press Finish A MAS file will have been automatically created within the new project It is also possible to start off with one of the many example project included in this plug in by selecting GOAL Example Project instead of GOAL Project in the first step above In the following screen select one of the example projects and press Finish to start working on it MAS files are files with extension mas2g When a new MAS file is created a MAS template is automatically loaded This template inserts the the sections that are required in a MAS file you should adapt the content of these sections of the MAS file to your needs A MAS file must have an agentfiles and launchpolicy section an environment section is optional After opening a new MAS file you should see something similar to Warnings are displayed because the sections are empty As can be seen warnings and errors are shown in at the line at which they occur underlining the specific text causing the issue In addition the Problems tab can be used to list all errors warnings Any errors need
3. W Print messages from warning 6 4 Runtime Settings Some settings e g used middleware are used as set in the preferences Note that these preferences can be set using the GOAL IDE but also by direct calls The raw settings yaml file that is used for persisting the preferences can also be edited directly Please refer to the JavaDoc for the various preference settings inside the goal preferences directory There are preference settings for the CorePreferences PMPreferences the Logging Preferences etc Also notice the w and v options of the tool if you are looking for warning printouts 6 5 Interaction with Standalone In some cases user interaction is still required in a stand alone run Usually this is because the environment needs additional setup actions before it can really be used Termination does not always imply reset of the environment You may have to set init param eters in the MAS to request reset of the environment for the next run 6 6 Batch Running The Run command above allows you various variants for batch running You can specify a directory containing multiple prepared MAS files and or you can request multiple runs of the MAS files enable proper batch running your setup has to meet two requirements e Your MAS terminates at some point This happens when all agents have terminated e g exited their main program section or were killed e Your environment resets the entities when it is killed and
4. 5 8 EXECUTING AN ACTION USING THE INTROSPECTOR 23 GOALDE File Edit Run Debug Sniffer 61 jeixiuia Process Overview bw2agents mas amp Blocks World Goals Mails Percepts Knowledge a tableagent agent stackbuilder stackbuilder me stackbuilder l agent tableagent jon g h jon e f on c d Z h Y c X g Z h Y b X g Z h Y table X g Z d 9 X c Z d c X c Z d b X c Z d table X c Z a 9 X b 2 Z a b X b Z a Y table X b Consolex Parselnfo Action log stackbuilder Database query assert belietbase2 a has the following solutions stackbuilder on b a has been added to the belief base and on b O has been deleted stackbuilder Database query beliefbase2 on a table on b a on c b on d table ta stackbuilder Executed move b a stackbuilder Database query true beliefbase2 on a table on b a b on d table ta stackbuilder Round 1 Figure 5 2 A sample query 5 3 Executing an Action using the Introspector You can also execute actions from the Query area within the Introspector The actions that can be executed include i mental state actions which change the mental state of the agent adopt drop insert delete and ii user specified environment actions t
5. To do this you provide a reference to the name of the already running environment e g env elevatorenv The name is derived from the jar file name and gets a serial number added to it if you launch multiple of them If you connect multiple MASs to a single environment the second MAS will only be able to connect to entities that are still free A GOAL multi agent system can be run in a truly distributed fashion having various agents running on different JVMs and or on different machines In this case there is one main platform that is the access point for the other platforms The main platform has to be launched first after which the other platforms can join Currently only RMI supports distributed running 4 6 DISTRIBUTED RUNNING OF ONE OR MORE MAS You should have two GOAL IDEs running note Eclipse not supported similar to Figure 4 3 and Figure 4 4 Note the remote process icon indicating that a process is running on a remote platform The IDE only allows introspection and debugging of local agent processes GOAL IDE File Edit Run Debug amp a Process Overview 9 elevator mas2g elevator elevator 3 elevator4 5 elevatorenv Figure 4 3 IDE for machine 1 GOAL IDE File Edit Run Debug Elal
6. if any and e Creating agents in line with the launch policy of the MAS 15 16 CHAPTER 4 RUNNING A MULTI AGENT SYSTEM How to launch a MAS e hight click a MAS file in the Script Explorer e Select Debug As GOAL Eclipse will automatically start the GOAL runtime when MAS is launched In this new window the Process panel is shown in the main content area as illustrated in Figure 4 1 The Process panel shows the processes that have been created by launching the MAS In the Blocks World example an environment process is started labeled Blocks World as well as two agent processes named stackbuilder and tableagent Also the Blocks World GUI shown in Figure 4 2 should appear All processes that are created are paused initially which is indicated by the icon just left of the process names Figure 4 2 The Blocks World GUI after launching bw2agents mas2g Before launching a MAS make sure to check that the MAS project is clean That is make sure no warnings or errors have been produced When you launch a multi agent system GOAL will not start the system when one or more program files contains an error Apart from the need to remove parsing errors it is also better to clean up program files and make sure that no warnings are present Please not that it also possible to run a MAS without a visual interface by selecting Run As GOAL instead of Debug As gt GOAL The Run functionality will run th
7. Lenvironment B BWT 2 BWAT2 BWATClient jar 3 insert proper initialization parameters and values on the next line or remove it T an mas2g 4 init clientip localhost clientport 2000 serverip localhost serverport 8000 oe 5 agentcount 3 humancount 0 launchgui false goal true humanbotgoal messagemod2g 7 robot goal Sagentfiles i messageTransiationp 9 X insert list of agent file references below HelloWorld 10 robot goal name robot 9 Test 9 WumpusAgents 15 LaunchpoLicy ha insert appropriate launch rules for your environment and agents below 15 when do launch robot 16 17 S Outine i ea reo 4 environment BWA4T2 BW4TClientjar init agentfiles robotgoal launchpolicy gt 3 v Problems 22 Console gt Oitems Description E Resource Path Location Type default package bw tmas2g BWAT2 Figure 3 3 Launch Policy of the BW4T example project 3 1 3 Adding a Launch Policy In order to create agents when a multi agent system is launched a launch policy for creating the agents is needed A launch policy is a set of rules which specify when to launch an agent A launch instruction creates and adds a new agent to the MAS when the MAS is launched A launch rule only creates an agent when a controllable entity in an environment becomes available i e when the GOAL platform is notif
8. environment ic lt s s 2o RR a k y eee 33 Ge POA a a uus a e E S Gt dr ee e arae 33 6 3 From the Command Line 0 05 2b da RR rA 33 BA Runtime L3 3o ee a X eee ea EROR REA 34 6 5 Interaction with Standalone 2e 34 Bi Boci RONDINE zs soe m Ae Qe ERG Sew Ohm BG 34 Creating Your Own Environments 35 7 1 Environments Distributed with GOAL 35 7 2 Creating an Environment Interface 35 Known Issues and Workarounds 37 Bl Reporting Issues ccs bk kh 9 ee RU ERE GOR ew a 38 Chapter 1 Introduction This user manual describes and explains how to e Install the GOAL platform e Start and use the GOAL plug in for the Eclipse IDE e MAS in stand alone mode without the IDE use this mode e g for running simulations and running the test framework and e Create your own environments that you can connect agents to using the Environment Inter face Standard EIS This document does not introduce the agent programming language GOAL itself For this we refer the reader to the GOAL Programming Guide 3 and GOAL s website 4 CHAPTER 1 INTRODUCTION Chapter 2 Installing the GOAL Platform This chapter explains how to install the GOAL platform We first describe the system requirements of the platform Please verify that your system meets the min
9. module s program section is empty 3 1 2 Adding an Environment to a MAS Project An environment is added to a MAS project by first adding it to the project which can be done for a new project or an existing one and then editing the environment section in a MAS file Note that the environment section is optional and that a multi agent system can be run without an environment Please note that you can also manually place the JAR file within the actual project location on your hard drive Eclipse will automatically update your project or right click the project and select Refresh to force a reload References to environments in a MAS file are resolved by checking whether the environment can be found relative to the folder where the MAS file is located or else an absolute path should be specified to the environment interface file The GOAL platform supports the Environment Interface Standard EIS to connect to environ ments 2 1 See Chapter 7 on how to develop an environment for creating your own environments Any environment that is compliant with the most up to date version of EIS that is supported can be used in combination with GOAL For connecting your agents to an environment that has been launched by another instance of the GOAL platform possibly on another machine check out Section 4 6 CHAPTER 3 CREATING AND MANAGING MULTI AGENT PROJECTS ane TIPO 12 Script Explorer S SSE DO bwat mas2g 7 4 Bwm
10. then restarted This may seem natural but in some configurations eg BW4T2 killing and restarting a local environment does not necessarily reset the real remote environment Chapter 7 Creating Your Own Environments 7 1 Environments Distributed with GOAL GOAL is distributed with several environments for which you can write your own agents These environments can be found in the GOALagents folder in the main GOAL folder in which GOAL is installed Running these environments is explained in Chapter 4 7 2 Creating an Environment Interface Environments provide so called controllable entities Agents can connect via an interface to these entities and control their behaviour The GOAL Programming Guide 3 explains in more detail how this works The interface used by GOAL is the Environment Interface Standard EIS 2 1 The current version of the GOAL platform is fully compliant with EIS v0 3 EIS is an Open Source project and documentation on how to create an interface for your environment such that it can operate in combination with GOAL is available via http sourceforge net projects apleis Ad ditional guidelines for creating an environment can be found on the website for GOAL http ii tudelft nl trac goal wiki Developers Environments HowToEISifyAnEnvironment 35 36 CHAPTER 7 CREATING YOUR OWN ENVIRONMENTS Chapter 8 Known Issues and Workarounds Below we have listed some known issues and possible workarounds when av
11. with an entity are only killed if the user kills it 4 4 Run Modes of a MAS and Its Agents A MAS or agent can be either in the running stepping paused or killed mode Controlling the run mode of a process is done with the Run P Step Pause 94 and Kill buttons in the tool bar Alternatively you can either use the Run menu or select a process and right click ctrl click on OSX to set the run mode You can control all processes that are part of the MAS at once by selecting the mas process node and using the relevant buttons This is particularly useful when starting stepping or killing all agents The following symbols are used to indicate the run mode of a process e indicates that the process is paused e indicates that the process is running e indicates that a process is in stepping mode 18 CHAPTER 4 RUNNING A MULTI AGENT SYSTEM e X indicates that the process has been killed Note It may happen that an agent tries to perform an action in an environment when the environment has been paused because the agent process and environment run asynchronously In that case the environment may throw an exception which will be printed as a warning in the console tab The agent will proceed as if the action has been succesfully executed Finally note that environments can be in running and paused mode but not in stepping mode 4 5 Resetting a MAS and Its Environment You can use reset to reset agents and environment If y
12. GOAL User Manual Koen V Hindriks Wouter Pasman and Vincent Koeman March 16 2014 Contents Introduction Installing the GOAL Platform 21 System Requirements o sa cy aa m RR Rue m Rom oe aem X E aoe n se ee 0 x o edo qoe ER A ae Ve oy 24 Uminstalling 42225224 a aa a ober os rode RUP DONE ke 250 ce ke A a E X a Sawa a Creating and Managing Multi Agent Projects 3 1 Creating New Multi Agent Project 3 1 1 Adding an Agent File toa MAS Project llle 3 1 2 Adding an Environment toa MAS Project 2138 Adding Launch Paley 9 3 1 4 Importing Knowledge and Module Files 3 2 Managing an Existing Multi Agent Project 3 2 1 Renaming Moving and 3 2 4 Automatic Completion a 2 2942 29 mo o EE ARRA Running a Multi Agent System 4 1 Launching a Multi Agent System 41 1 Initializing an Environment 2223293 om Ray s 4 2 Running a Multi Agent System ee ee 4 3 Terminating a MAS and or Its 44 Run Modes of a MAS and Its 4 5 Resetting MAS and Its
13. Parselnfo Action log stackbuilder tableagent Initializing mental state of agent stackbuilder gent stackbuilder loaded successfully Initializing mental state of agent tableagent gent tableagent loaded successfully Launching Scheduler Scheduler initialized Figure 5 7 Export dialog for mental state contents 5 6 2 Prolog Exceptions When an exception occurs in the Prolog engine SWI Prolog error messages may get a bit hard to decipher This section explains how to read such messages A typical PrologException will show up typically like this WARNING jpl PrologException PrologException error instantiation error context system gt 2 _1425 query true beliefbase6 openRequest0nRoute L D T The first of the two lines indicates the message that Prolog gave us The second line shows the query that caused the warning To read the first line notice the part error CAUSE DETAILS inside it The CAUSE part gives the reason for the failure the DETAILS gives the context of the error Several types of CAUSE can occur amongst others 1 type_error X Y an object of type X was expected but Y was found For instance in the context of arithmetic X can be evaluable and Y a non arithmic function 2 instantiation error the arguments for an operator are insufficiently instantiated For exam ple if we ask X gt 8 with X a free variable we will get an error like the example above
14. The context in such a case would be context system gt 2 1425 The first part system 2 refers to the gt operator which has arity 2 and is part of the system module and the 1425 refers to the free variable which apparently was even anonymous hence is not of much help Note that these messages commonly contain pure functional notation where you might be more used to the infix notation For instance system X appears as system X and gt 2 appears as gt 2 3 representation error the form of the predicate was not appropriate for instance a bound variable was found while an unbound variable was expected 5 7 RUNTIME PREFERENCE SETTINGS 31 4 permission error you have insufficient permissions to execute the predicate Typically this occurs if you try to redefine system predicates 5 resource error there are not enough resources of the given type 5 7 Runtime Preference Settings Most settings can be modified from the preferences in Eclipse Access them through opening the GOAL category in Window Preferences and selecting the Runtime page This section discusses those settings 5 7 1 Platform Management The tab PlatformManager in the Help Preferences menu offer a number of options to set the base path for the environment and middleware This preference window offers the option to always run the middleware locally This checkbox is by default disabled and therefore when starting a mas you will a
15. a e oT 2 je x mja gt wm X s j Process Overview elevator2 mas2g amp elevator 9 elevator_3 9 elevator 4 elevatorenv Actionlog elevator 3x _ elevator 4 x Received new entity named car2 initializing mental state connecting to entity car2 OK 88 milliseconds to load MAS trying to create registry 4 Figure 4 4 IDE for machine 2 If you take down the second MAS the entities will be released and can be reused by other still running MASs If you take down the MAS that launched an environment the environment will be taken down removing all associated entities All agents associated to these entities will then be killed 20 CHAPTER 4 RUNNING A MULTI AGENT SYSTEM Chapter 5 Debugging a Multi Agent System The GOAL runtime provides various options for debugging While running a mas GOAL provides a range of options for debugging which include the following 1 options for viewing and controlling debug output produced while running a mas 2 setting general and specific breakpoints to pause an agent on 3 agent introspectors for inspecting the mental state of an agent and percepts and messages received at runtime 4 a message sniffer for tracing messages that are exchanged between agents currently only when running the JADE middleware TODO Goat facilitates debugging of a multi agent syst
16. able Console Parselnfo Action log stackbuilder Database query assert belietbase2 a has the following solutions 1 stackbuilder on b a has been added to the belief base and on b c has been deleted stackbuilder Database query beliefbase2 on a table a on c b on d table table stackbuilder Executed move b a stackbuilder Database query true beliefbase2 on a table on b on c b on d table table stackbuilder Round 1 Figure 5 1 Introspector for the stackbuilder agent 5 2 Querying a Mental State Apart from being able to view the contents of an agent s mental state when it is paused an introspector also allows you to pose queries for evaluation on the current mental state of the agent in the Query Area All solutions for the query performed are shown in the query window They are shown in the query result area as a list of substitutions for variables Figure 5 2 shows the results of performing the query bel clear X clear Y on X Z not on X Y This mental atom corresponds to the precondition of the move action in the Blocks World environment and shows all the options to move blocks that are available in the present state Consult the GOAL Programming Guide 3 for a detailed explanation of mental state conditions
17. ailable Nothing happens when pressing the Step and Run buttons There are a number of possible explanations for this you can check 1 First make sure that the environment is ready to run For example when using the Elevator environment the environment will respond to GOAL only if you selected GOAL Controller available in this environment 2 You may be stepping an agent that is suspended somehow e g the environment may be refusing to serve that agent or it may be the turn of another agent The environment does not work as expected Check the documentation that comes with the environment Most environments need to be set up before the agents can connect and perform actions The Wumpus environment hangs when the agent shoots an arrow The Wumpus World always has to be enclosed by means of walls If an agent shoots an arrow into open space the arrow will continue forever causing the system to hang The Reset button does not work It is not always possible to reset an environment as an environment may not provide the support for doing so I do not see an environment after launching a MAS The environment may be hidden behind another window This is due to a known bug in Java for Windows You need to install Java 7 or higher to fix this issue Java 7 is available from http download java net jdk7 binaries 37 38 CHAPTER 8 KNOWN ISSUES AND WORKAROUNDS 8 1 Reporting Issues If you discover anything you believe to be an erro
18. akpoints preferences panel Figure 5 3 The warnings print format allows you to select the information that is printed along with any warning You can turn on the stack trace and the detailed java message Also you can set the maximum number of times the same message is printed to avoid flooding of the console 5 5 Setting Breakpoints A useful way to trace what your agents are doing is to step the multi agent system and check how and which action rules are evaluated which actions are performed which percepts are received etc Stepping assumes natural breakpoints to stop an agent at For GOAL agents a number of such breakpoints have been pre defined These built in breakpoints facilitate stepping and tracing of what happens when an agent is executed A second alternative method to set breakpoints is by explicitly adding them to code lines in agent program files A user can set breakpoints at various places in a program where the agent should stop when such a breakpoint is hit 5 5 1 Stepping and Global Breakpoints The first method to step an agent and halt at breakpoints is based on a set of built in breakpoints that relate to the reasoning cycle of an agent A list of these breakpoints is available in Eclipse by opening the GOAL category in Window Preferences and selecting the Logging page 5 3 For each breakpoint a selection box with multiple options is available If View Only is selected a report related to the breakpoint will
19. be made available in the Debug Tracer of an agent If View and pause is selected in addition the creation of a report all agents will be paused when such a breakpoint is hit The Do nothing option skips the breakpoint completley We briefly discuss each of the available built in breakpoints e The first item Steps in the reasoning cycle of an agent relates to e g the start of the agent s reasoning cycle By default a round separator is added to the output in the Debug Tracer 26 4 GOAL CHAPTER 5 DEBUGGING A MULTI AGENT SYSTEM GOAL Eclipse 5 omm R o Quick Acce El amp java 9 General Android Ant Code Recommenders Default KR language CodePro Sleep agents when they do no actions Dynamic Languages Remove agents when they are killed V Open debug trace tab when agent is launched Prefix agent names with MAS name V Print message when new entity appears L New agents copy the environment run state or run if no environment is available Always run the middleware locally Enable learning Learned behaviour file Browse debug messages Do nothing The reasoning cycle separator View and pause Metrics Preferences Mails received View only Mylyn Plug in Development Prolog Entry of the init event or main module Do nothing Run Debug Server Team Rule application summary View only Percepts processed View only Entry of a user defined module Do nothing Evaluation of rule conditions Do nothin
20. e MAS through a console command line which is useful for quick testing of the desired functionalities 4 1 1 Initializing an Environment In order to actually run a MAS it may also be necessary to initialize an environment by means of the environment s user interface In the UNREAL TOURNAMENT environment for example you can create new bots that can be controlled by agents Please consult the environment documentation for details Depending on the configuration and launch rules of your MAS file new agents are automatically created when a new bot is added Refer to the Programming Guide for more information on this 3 Note that environments often also have entities that cannot be controlled by agents such as people taking the elevator in the Elevator environment or UNREAL bots that are controlled by the UNREAL engine itself 1 status of the environment as indicated by the icon next to the environment process name may not always be known In that case the icon may be shown to indicate the state is unknown This icon will be shown with some environments that do not provide information about their current state 4 2 RUNNING A MULTI AGENT SYSTEM 17 4 2 Running a Multi Agent System After an environment has been initialized a multi agent system can be run How to run a MAS e Select the MAS node in the Process panel e Press the Run button to run the multi agent system processes including the environment and the age
21. ea This tab typically contains more detailed information about your agent avoid flooding the tab it is only filled however when the multi agent system is paused To be useful the contents of the tab need to be analyzed which cannot be done while new content would be added continuously The exact contents of the tab can be customized by means of the Help Preferences Breakpoints settings menu Various items that are part of the reasoning cycle of an agent can be selected for viewing If checked related reports will be produced in the debug tracer of an agent more detailed explanation of these breakpoints can be found in Section 5 5 If you step a multi agent system using the Step button 4 the runtime by default switches to the debug tracer of the agent that is being stepped This allows you to automatically keep track of what happens without having to manually switching between agent debug tracer tabs The 5 5 SETTING BREAKPOINTS 25 default switching between debug tracers can be turned of by means of the Help Preferences GUI menu uncheck the option at the bottom labeled Automatically switch to Debug Tracer tab Debug tracer tabs can be closed by clicking the close icon on of the tab You can re open a closed tab by selecting the agent in the process panel and righ clicking ctrl click on OSX and selecting Open Debug Tracer Customizing the Logging The logging behaviour of the runtime can be modified from the preferences in Ecl
22. em in various ways It provides various forms of feedback in the feedback area at the bottom of the runtime in debug mode i e when a multi agent system is running In addition for each agent an introspector can be opened to inspect the contents of the mental state of an agent Finally it is possible to set breakpoints to pause an agent when it hits such a breakpoint 5 1 Introspecting Agents A useful tool to trace what your agents are up to is the introspector An introspector allows you to inspect the mental state of an agent How to open an Introspector e Open an introspector by double clicking on an agent process in the Process panel or e Alternatively select the agent process and the Open Introspector item in the Debug menu Introspectors are located in the content area of the debug mode Figure 5 1 shows an intro spector for the stackbuilder agent that is part of the bw2agents mas2g project An agent introspector consists of various tabs that allow you to access the current contents of the agent s beliefs goals mails percepts and knowledge In the area showing the contents you can select text and copy it to somewhere else 21 22 CHAPTER 5 DEBUGGING A MULTI AGENT SYSTEM Elejgialgiw jeixiuwja gt gt 5 B 4 Process Overview bw2agents mas Blocks World tableagent agent stackbuilder Eee me stackbuilder agent tableagent on g h on e f on c d on f t
23. eparate file that can be imported into Prolog Then use the Prolog tracer to analyze the code 5 6 1 Exporting a Mental State to a File To export the mental state of a GOAL agent select the agent process and then select Debug Export mental state This allows you to save the contents of a mental state as a Prolog file The exported file s can then be imported into a Prolog debugger SWI Prolog can be used for this provided that you installed SWI Prolog on your machine The version of SWI Prolog that is distributed with GOAL cannot be used for this purpose A detailed set of instructions on how to do this follows 1 select and pause the agent just before the point where you would like to debug a call to Prolog 2 select menu Debug Export Mental State contents and save the database contents as a ms MentalState file You can save beliefs and goals to a separate file and opening the exported file in the IDE but for opening in Prolog export to a single file seems most useful After clicking OK a file browser appears select a good place to save the file 3 Manually edit the ms files a They need to have the extension 1 to import in prolog b predicates that you use but do not have explicit values in the prolog program may need to be declared dynamic See the SWI Prolog Manual for details c Goals are dumped to the ms files as a conjunct of predicates and this is not a legal notation in Prolog the least you have to do
24. file in Eclipse and adding breakpoints at the first line associated with the construct That is put a breakpoint next to the line with the module name the first line associated with an action rule or the the line with the action name You can do this by double clicking in the left border of the editor GOAL will automatically allocate placed breakpoints to their appropriate position when the MAS is started by attaching any placed breakpoint to the first suitable location after its current position There are two types of breakpoints the ALWAYS and the CONDITIONAL type A CONDI TIONAL breakpoint shows as an orange icon in the left border It can only attach to the action 28 CHAPTER 5 DEBUGGING A MULTI AGENT SYSTEM part of an action rule The ALWAYS type shows as a red icon in the left border and attaches to the other three types An example of two breakpoints that were added to the move action specification is illustrated in Figure5 5 fL GONL ewtoyeevaomgentgos Edipge 0 0 8 1 file Edit Source Navigate Search Project CodePro Run Window Help mancen l 1 Scnpt Explorer es C Q9 elevatoragent goat ea 9 ewm au 15 H elevator 16 goais elevatormasg 1 atFloor 1 dir up elevatoragent goal 18 elevatorenvjar 19 actionspec Helloworid 2 The action Mr 22 atFloor Her 9 Wornpushgents 2 X When the elev goto Level Dir 5 pre atFloor Here dir DirNow post not dir DirNo
25. g WindowBuilder Evaluation of user specified action preconditions View only XML Built in actions that have been executed Do nothing User specified actions that have been executed only Changes to the belief base Do nothing Changes to the goal base Do nothing Goals that have been achieved View only Results of mental atom queries Do nothing Resource Path Location Type Restore Defaults Apply Cancel Figure 5 3 Setting Global Breakpoints to indicate which reasoning cycle round has been executed This round separator is also available in the action log The items Mails received and Percepts received concern the events of receiving new mails and percepts Mails and percepts are buffered and collected at the start of an agents reasoning cycle A report containing the percepts and or messages received is provided as output in the Debug Tracer Evaluation of action rules concerns the evaluation of the mental state condition and appli cability of action rules Depending on the rule evaluation order more or less rules will be evaluated each reasoning cycle The debug output will show all substitutions for which the rule is applicable if any Evaluation of action preconditions relates to the evaluation of the preconditions of actions In case this breakpoint is selected to pause on the agent will halt just before executing the action itself If this breakpoint is viewed reports of the evaluation of all action preconditions are
26. gent System This chapter explains how to launch run and distribute a MAS project across multiple machines Running a multi agent system means launching or connecting to the environment the agents perform their actions in if any and launching the agents when appropriate As explained in Section 3 1 3 which environment is launched depends on the environment section of the MAS file that is being run Which agents are launched depends on the launch policy specified in the same MAS file Section 4 6 explains how to distribute one or more MAS across multiple machines Please not that although currently the GOAL runtime is launched in a separate window we are working on fully integrating this in Eclipse GOAL IDE GOAL IDE File Edit Run Debug 2 4 Edit Process Overview bw2agents mas2g blocksworld blocksworld stackbuilder tableagent gt Console x ParseInfox Actionlog stackbuilder x tableagent x Took 73 milliseconds to load MAS initializing mental state connecting to entity gripper OK Welcome to the blocks world Click blocks to stack Spacebar to reset initializing mental state connecting to entity gripper OK rire lt gt Figure 4 1 The runtime after launchingbw2agents mas2g 4 1 Launching a Multi Agent System Launching a MAS means e Launching an environment referenced in the MAS file
27. hat are executed in the environment You cannot execute other actions such as combined actions sending messages or modules Make sure that the action is closed i e has no free variables To execute an environment action make sure that the environment is running You may also need to figure out whether the 24 CHAPTER 5 DEBUGGING MULTLAGENT SYSTEM agent that should execute the action is allowed to perform an action at the time you want to perform the action e g it is the turn of that agent Note that an environment action is sent to the environment without checking its preconditions The postcondition of the action also is not processed nor are any percepts related to the action passed to the agent Finally no feedback is provided whether an action was successfully performed or not 5 4 Debugging Output In debug mode the feedback area in the runtime contains various tabs for inspecting agents and the actions they perform see Figure 4 1 Besides the console tab an action log tab is present that provides an overview of actions that have been performed by all running agents In addition when an agent is launched a debug tracer tab is added to the feedback area for inspecting various aspects of the agent program during runtime It is sometimes useful to clean the contents of a tab This can be done by right clicking ctrl click on OSX and selecting the Clear option This is particularly useful for the action log which is ne
28. ied that an entity has been born Agent name and agent file name can but do not have to be different Finally more than one agent can be connected to a single entity The GOAL Programming Guide 3 provides a comprehensive explanation of launch policies 3 1 4 Importing Knowledge and Module Files When you start writing larger agent programs it is often useful to add more structure to the MAS project GOAL provides the import command to import contents from other files into an agent program important reason for moving code to separate files and importing those files is to facilitate reuse 3 2 MANAGING AN EXISTING MULTI AGENT PROJECT 13 A procedure very similar to that for knowledge files can also be used to move complete modules to separate module files with extension mod2g These are the two types of files that can be imported Files that contain modules and files that contain knowledge Even if an agent is not a very large program it sometimes is useful to put some of the code in separate module or knowledge files to facilitate reuse The code in knowledge section of an agent for example can be re used in another agent program Reusing the same code by creating a separate file that is used in multiple agent files prevents possible problems introduced by modifying the same code that appears at different places in only one location By using a dedicated file the code is needs to be maintained and modified in only one
29. imal system requirements 2 1 System Requirements The GOAL platform runs on Windows Macintosh OSX and the Linux operating system For information on the exact versions of these OSs please check GOAL s website at http ii tudelft nl trac goal The GOAL platform requires Java SUN or OpenJDK version 1 6 or higher We recommend using Sun Java 1 7 2 2 Installation GOAL is available as a plug in for Eclipse Kepler see https www eclipse org downloads It can be installed as follows If you are a Linux user please make sure to run Eclipse as administrator at all times On Windows if User Account Control is enabled Eclipse itself should either be installed in a user owned folder or run as administrator 8 CHAPTER 2 INSTALLING THE GOAL PLATFORM ELLOS S Outine 1 An outline is not available Protsems Console 11 Android Figure 2 1 The GOAL perspective in Eclipse 2 3 Uninstalling For uninstalling the GOAL plug in either remove your whole Eclipse installation or follow these steps to uninstall just the plug in 2 4 Upgrading Upgrading the GOAL plug in to a newer version outside of the aforementioned automatic update functionality is possible in Eclipse by selecting Help Check for Updates 2 5 Customizing Various GOAL platform features can be customized e g debugging features The default settings can always be restored using the Restore Default button
30. ipse Access them through opening the GOAL category in Window Preferences and selecting the Logging page A timestamp with millisecond accuracy can be added to each printed log by checking the Show log time checkbox The write logs to file checkbox determines if GOAL writes all logs to separate files Logs for each agent the action log warnings results from GOAL log actions etc are all written to separate files Log files are written to the logs directory inside the directory where GOAL was installed You can change it by clicking on the Browse button of the Write log files here field Most logs are in XML format and each log entry contains a timestamp with millisecond accu racy Debug messages for each jagent are written to lt agent gt log actions are written to histo ryOfActions log General info is written to goal tools logging InfoLog log Warnings are written to goal tools errorhandling Warning log Sytem bug reports are written to goal tools errorhandling SystemBug log Result of log actions are written to agent timestamp txt files Be careful with this option as many megabytes per second can be written A new run generates new log files made unique by adding a number to the names By selecting Overwrite old log files a new run uses the same file names thus overwriting previously generated files The type of debug messages that are logged can be determined with the view checkboxes in the global bre
31. is to replace all commas within a single goal with dots d You may need to fix the goals Several goals may be in conflict when seen from SWI Prolog and therefore you may have to remove part of the goals 4 open SWI Prolog see the SWI Prolog manual and import the Prolog file e g consult lt filename gt 5 start the SWI Prolog tracer e g type trace in its console 6 run and trace the query see SWI Prolog manual typically the n a and enter are used for tracing If you selected opening of files while exporting the mental state contents these files are opened in edit mode Switch back to edit mode in order to view the contents Figure 5 7 shows the results for the stackbuilder agent 30 CHAPTER 5 DEBUGGING A MULTI AGENT SYSTEM amp GOAL IDE File Edit Run Debug Sniffer Help memes aexne iles goals pl P MAS Projects beliefspl X M coffee mas2 Ee owledge lr 2 block X f 8 clear X block X not on X EN ii coffeegrinder goal 4 clear table T bw2agents 5 tower X on X table stackBuilder goal tover X YIT on X Y tower IT 6 tableAgent goal Es beliefs E 8 blocksworld agent stackbuilder 6 stackBuilder goal 9 ne stackbuilder E Other files 10 agent tableagent percepts Console x
32. just the number of times that a message will be re printed in the Logging tab in the Preferences menu Chapter 6 Running in Stand Alone Mode You can run a multi agent system containing GOAL agent programs stand alone that is without starting the IDE or Eclipse That way you can use GOAL from scripting languages like a shell script or bat script or from your Java program Stand alone running always involves starting up a Java Virtual Machine JVM with the right environment settings We first describe the required environment and then describe how to use this from a Java program or from a script 6 1 Setting the JVM environment The variables to be set up properly for the environment are the PATH the JAVA LIBRARY PATH CLASSPATH and the SWI HOME DIR Below we use the tag GOALHOME to point to the absolute directory where you installed GoAL The exact way to do this is machine dependent we assume you have the knowledge how to do this on your particular machine For example the used as separator in the CLASSPATH below should be replaced with a on Linux and OSX e on Windows PATH should be set to include GOALHOME swifiles libs On Linux you should set LD_LIBRARY_PATH instead and on OSX you should set DYLD LIBRARY PATH e SWI HOME DIR should be set to point to GOALHOME swifiles e JAVA LIBRARY PATH should be set to include GOALHOME swifiles libs e CLASSPATH should be set to include GOALHOME goal jar and to the EIS en
33. kbuilder Evaluating event module stackbuilder Rule forall bel percept on X Y not on X Y do insert on X Y is not applicable stackbuilder Rule forall bel on X Y not percept on X Y do delete on X Y is not applicable stackbuilder Evaluating main agent program stackbuilder Rule if misplaced X then move X table has 1 potentially applicable instantiation s X c stackbuilder Action move c table has no available instantiations stackbuilder Rule if misplaced X then move X table is not applicable after all stackbuilder Rule if constructiveMove X Y then move X Y has 1 potentially applicable instantiation s Y b X c stackbuilder Precondition clear c clear b on c Z not on c b of action move c b holds when instantiated by Z table stackbuilder Goal on a table on b a on c b on d table on e table on f e on g f on h g on i h has been achieved and is removed LiT Ji Figure 5 4 Pausing on a Goal Achieved Breakpoint 5 5 2 Setting Code Line Breakpoints second more flexible way of setting breakpoints is by adding code line breakpoints In GOAL breakpoints can be associated with four constructs in an agent program a module the condition of an action rule the action part of an action rule and an action specification Breakpoints can be added to these constructs by opening a mas project
34. lways be prompted for the location of the middleware host If you turn on this checkbox the default localhost will be used as middleware host and the dialog will not be shown You can adjust the directory where browsing for mas2g and goal files will start and whether your last directory should be used as the starting point for browsing The default KR Language selection allows you to choose between the default swiprolog imple mentation and an alternative swiprolog2 implementation swiprolog2 attempts to compile mental state queries into a single query to the underlying swi prolog engine This implementation is not fully tested and it is recommended to use swiprolog Selecting the sleep agents when they repeat same action enables GOAL to skip agents that seem to be idle which relocates CPU time to other agents that really are working on something GOAL guesses that an agent is idle if the agent is receiving the same percepts and mails as last step on the input and picks the same actions as output The sleep is cancelled when the percepts mails of the agent change This setting is on by default IMPORTANT with sleep agents selected your agent may go to sleep when it is repeating actions This may be counterintuitive and may cause GOAL programs to malfunction In doubt you should turn off this option To remove agents automatically after they have been killed select the Remove agents when they are killed option To run new age
35. nts must be in running mode to be able to run the complete multi agent system The symbol indicates that a process is in running mode In running mode an agent will execute a complete reasoning cycle without pausing If the agents in a MAS are paused but the environment is not the environment still may be changing due to processes that are not controlled by the MAS or simply because actions take time durative actions Vice versa if an environment is paused but one or more agents in the MAS are running typically you will see complaints that the agent was not able to retrieve percepts from the environment 4 3 Terminating a MAS and or Its Agents How to terminate a MAS or an Agent e Select the MAS or agent you want to kill in the Process panel e Press the Kill button in the tool bar Alternatively press the red stop button within Eclipse or just quit the runtime window The GOAL kill button is preferred though Terminating a MAS terminates the environment and all agents Killing a single agent does not kill the MAS The X symbol indicates that a process has been killed An agent that has been killed is not performing any actions any more If a killed agent is put back in running mode the agent is restarted and initialized using the initial mental state specified in the agent program agent will also be automatically killed if the entity that it is controlling in an environment has been terminated Agents that are not associated
36. nts automatically as their entities appear select Run agents automatically when the environment is already running These two options are useful in environments where large numbers of agents appear and disappear while GOAL is running 5 7 2 Warnings Warnings and errors are shown depending on how you set the preferences Warnings generally appear in the Console area see Section 5 4 By default warnings are plain messages For development purposes however the format of warnings can be adjusted using the Logging tab in the Preferences menu Select to Show Java details to include the Java specific part of a warning message In some cases this can help to see what the platform was attempting to do Select Show Java stack traces to add a full stack dump to warning messages This is mainly intended to aid debugging of the platform itself but in some cases may also help you to understand what the platform was trying to do Reporting Issues When reporting an issue it is very useful if your issue report includes Java details and stack traces See Section 8 1 for more about reporting an issue 32 CHAPTER 5 DEBUGGING A MULTLAGENT SYSTEM When an identical problem re occurs a warning message is re printed only a limited number of times to prevent the Console from being flushed with identical messages When the Console shows the message for the last time it will add the message New occurrences of this warning will not be shown any more You can ad
37. ou select MAS process node and apply reset only the agents are reset but not the environment To reset the environment select the environment and apply reset Note that reset of the environment keeps the agents connected and alive so this is less complete than a kill and restart of the entire MAS The Backward stepping button in the tool bar is not yet supported 4 6 Distributed Running of One or More MAS The agents in a multi agent system and the environment to which the agents are connected can be distributed over different machines The middleware infrastructure that is used to run the multi agent system needs to support distributed processing 4 6 1 Messaging Infrastructures The term messaging refers to the communication infrastructure on top of which a multi agent system runs Messaging enables agents to exchange messages between each other GOAL currently provides two alternative messaging platforms Local messaging and a messaging based on RMI The local middleware is selected by default since the performance of the local middleware usually outperforms that of RMI But to run a truly distributed agent system you need to select RMI Messaging Use the Run menu to select one of these 4 6 2 Distributing MAS It is possible to connect a MAS to an existing environment that is already running in another instance of the GOAL platform This can happen for example when a second MAS is run using a second instance of the GOAL platform
38. place 3 2 Managing an Existing Multi Agent Project Loading an existing multi agent system project can be done through the following steps 14 CHAPTER 3 CREATING AND MANAGING MULTI AGENT PROJECTS Please not that only files that are in the sub directory of the selected mas2g file will be copied or directly used in the new project Any external files will have to be moved into the project manually by using the file system or the steps as explained in Section 3 1 2 3 2 1 Renaming Moving and Deleting a File The name of a file can be changed by right clicking on it and selecting Refactor Rename Note that a file can also be moved through the Refactor menu Deleting a file can simply be done by right clicking on it and selecting Delete A warning is in place however Changing a file name or location that is part of a MAS project does not modify the contents of the mas file itself When changing an agent file name or moving or deleting it you need to manually change the agentfiles section of the mas file All of these operations can also be done on the file system itself 3 2 2 Automatic Completion Automatic code completion is fully supported Using auto completion e After typing some initial code it can be automatically completed by pressing CTRL Space e If there is more than one option for completion select an option from the pop up menu otherwise the word is completed right away Chapter 4 Running a Multi A
39. program stackbuilder Rule if misplaced X then move X table has 9 potentially applicable instantiation s X a X d X e X c X f X g X R X i istackbuilder Precondition clear h clear table on h Z not on h table of action move h table holds when instantiated by Z i D 4 il I D gt Figure 5 6 The stackbuilder agent paused on an explicit breakpoint Breakpoints that are associated with action specifications and action rules provide useful output in the Debug Tracer immediately This is not the case however for breakpoints set on modules The output produced in this case only provides a clue that the module is about to be evaluated Take it from here to step further through the module using the Step button to produce more relevant debug output 5 6 PROLOG LEVEL DEBUGGING 29 Explicit breakpoints that have been added to a file are not permanently saved Upon closing Eclipse all breakpoints set are lost This is an easy way to clear all breakpoints Individual breakpoints can be cleared by double clicking them again Eclipse will toggle between the two different types of breakpoint and no breakpoint at all 5 6 Prolog Level Debugging GOAL does not support tracing of Prolog queries inside the Prolog inference engine If you need this level of detail while you are debugging an agent it is useful to be able to export the contents of a mental state to a s
40. provided Besides the action log which actions are executed can be provided as debug output in the Debug Tracer as well by checking the View check box related to either the Built in actions or Actions sent to the environment breakpoints The Changes to the belief base and changes to the goal base breakpoints provide output related to or pause when a change to either base is made The Changes to the belief base breakpoint usually is a too fine grained breakpoint to pause on as many changes to a belief base are made The Goals that have been achieved breakpoint can be quite useful to verify whether a goal has been achieved and to halt an agent whenever this happens The last two breakpoints are the most finegrained level breakpoints available and typically are not very useful They provide detailed output on basic queries at the level of mental atoms and or queries performed using the knowledge representation language e g Prolog 5 5 SETTING BREAKPOINTS 27 You can modify the selections at any time also during the execution of a mas If you are stepping a multi agent system agents will continue to run until they hit the next breakpoint selected to pause on If such a breakpoint is hit the agent is switched to pause mode Note that the Step button has the same effect as the Run button if no breakpoints have been selected In that case the agent progress will be run without halting the agent with the exception of explicitly set breakpoin
41. r or have any other issues you run into please report the error or issue to goalQii tudelft nl Your help is appreciated To facilitate a quick resolution and to help us reproduce and locate the problem please include the following information in your error report 1 A short description of the problem 2 relevant source files of the MAS project 3 Relevant warning or error messages from the Feedback area especially from the Console and or Parse Info tab Please also include Java details and stack trace information if avail able see Section 5 7 2 for adjusting settings to obtain this information 4 If possible a screen shot of the situation you ran into Bibliography Tristan Behrens Environment interface standard project http sourceforge net projects apleis Tristan Behrens Koen V Hindriks and Jrgen Dix Towards an environment interface standard for agent platforms Annals of Mathematics and Artificial Intelligence pages 1 35 2010 10 1007 s10472 010 9215 9 Koen V Hindriks GOAL Programming Guide Delft University of Technology 2014 http ii tudelft nl trac goal wiki WikiStartitDocumentation Koen V Hindriks The GoAL Agent Programming Language http ii tudelft nl trac goal 2014 39
42. to be fixed before we can launch a MAS Note that multiple MAS files can be added to a single project e g for different run configurations Save a file either by using the Save icon in the menu or by pressing CTRL S Windows and Linux or APPLE S OSX 9 10 CHAPTER 3 CREATING AND MANAGING MULTI AGENT PROJECTS GOAL Edipse ON File Edit Refactor Navigate Search Project CodePro Run Window Help DILE LLL 12 Sonpt Explorer SS en o HelloWorid 9 WumpusAgents Create new GOAL Project Create a new GOAL Project in the workspace Project name MyProject Environment optional _C Users Vincent GOAL GOALagents HelloWorld Hi Outline 7765 An outline is not available Console alME n 7n5 Android Test Figure 3 1 The New Project Dialogue Window Edit Source Refactor Navigate Search Project CodePro Run Window Help ok ie IR Mata een Senpt Explorer 1 a SO Test mas2g 2 eo ewm Lagentfiles Helioworid 2 4 Test 2 Tens pA agentfiles s launchpolicy 7 WumpusAgents Outline 11 M agentfiles launchpolicy E Console 12 Android default package Testmas2g Test Figure 3 2 A new MAS File 3 1 CREATING NEW MULTI AGENT PROJECT 11 3 1 1 Adding an Agent File to a MAS Project A new agent file is automatically loaded with a corresponding template Again warnings are displayed for a new file because the main
43. ts which we discuss next As an example clear all Pause check boxes in the Breakpoints panel and only select to pause on the Goals that have been achieved breakpoint Launch the bw2agents mas2g mas and press the Step button several times to get things started The agents run until the stackbuilder agent achieves its goal which demonstrates the agent is able to achieve it even in the presence of the tableagent agent Figure 5 4 shows an example run where this happens in round 20 amp GOALIDE File Edit Run Debug Sniffer Help bw2agents mas2g E Sackbuker x Blocks World Beliefs Goals Mails Percepts Knowledge tableagent n a table on b a on c b on d table on e table on f e on g f on h g on i h ar stackbuilder Action move b table has no available instantiations stackbuilder Rule if misplaced X then move X table is not applicable after all stackbuilder Rule if constructiveMove X Y then move X Y has 1 potentially applicable instantiation s Y a X b stackbuilder Precondition clear b clear a on b Z not on b a of action move b a holds when instantiated by Z table stackbuilder Round 20 stackbuilder Received percepts percept on a table percept on b a percept on c table percept on d table percept on e table percept on f e stac
44. ver automatically cleared Console The console shows all important messages warnings and errors that may occur These messages include those generated by GOAL at runtime but possibly also messages produced by environments and other components When something goes wrong it is always useful to inspect the console tab for any new messages If a new message is present in the console tab the color of the tab is changed to indicate this Action Log The action log tab shows the actions that have been performed by running agents that are part of the multi agent system These actions include both user defined actions that are sent to the environment as well built in actions provided by GOAL Apart from an overview of what has been done by agents the action log can also be used as a simple means for debugging It can be used for example to check whether a particular action added for debugging has been performed The idea is that you can add a dummy action with a particular content message to a rule and verify in the action log whether that action has been performed To do so include an action in the action specification of your agent named e g debug lt yourMessage gt with true as pre and post condition to ensure the action always can be performed and will not change the state of the agent but still will be printed in the action log output Debug Tracer Upon launching an agent a debug tracer tab is added for that agent in the feedback ar
45. vironment jar file that you want to use You can inspect the goal bat Windows or goal sh file Linux OSX to see a running example of the above 6 2 From Java Running a stand alone MAS from Java is very simple just create the goal tools Run object from Java and start it using this call new Run new File fileName run where fileName is the MAS file that you want to run 6 3 From the Command Line To run GOAL stand alone from the command line the typical call looks like this 33 34 CHAPTER 6 RUNNING IN STAND ALONE MODE java cp goal jar Djava library path swifiles libs goal tools Run lt filename mas2g gt wait where lt filename mas2g gt is the absolute path to the MAS file you want to run You can add the wait option so that the runner waits between starting the environment and scheduling the GOAL agents see Section 6 5 below d debug Display out put from debugger while running agent Print messages from error h help Displays this help 1 Print messages from info license Shows the license p Print messages from parser jnumber j Number of times to repeat running all episodes recursive Recursively search for mas files rmi jhostj Use RMI messaging middleware Host is the location of the RMI server Using localhost will initialize a RMI server timeout jseconds Timeout of the multi agent system v verbose Print all messages version Shows the current version
46. w dir Dir 28 29 31main module 2 program Goto the nearest floor that the agent wants to go to if a goal atFloor Level dir Dir 35 bel atFloor Here Distance is abs Level Here St Outine 247 10 not a goal atFloor OtherLevel OtherDistance ts abs OtherLevel Here OtherDistance lt Distance 4 a initmoduled initmodule then goto Level Dir 9 knowledge beliets 40 goals 41 actionspec i2event module 4 4 mainmoduleQ 3 program program SRK Percept Processing v 4 a eventmodule program I Problems T 0 items Description 5 Resource Path Location Type Figure 5 5 An ALWAYS and a CONDITIONAL Breakpoint added to theelevator Different from the built in breakpoints even if a multi agent system is put in run mode the system will halt on any explicit breakpoints that have been set Figure 5 6 shows that the stackbuilder is paused when the breakpoint set on the action specification is hit 5 GOAL IDE Te File Edit Run Debug Sniffer Help rocess Overview bw2agents mas2g Blocks World y tableagent stackbuilder D istackbuilder Executed insert on f table stackbuilder Executed insert on h i stackbuilder Executed insert on i table stackbuilder Rule forall bel on X Y not percept on X Y do delete on X Y is not applicable LI stackbuilder Evaluating main agent
Download Pdf Manuals
Related Search