Introduction to AgentSpeak and Jason for Programming Multi-agent Systems (2)

1. Introduction to AgentSpeak and Jason for Programming Multi-agent Systems(2) Dr Fuhua (Oscar) Lin SCIS Athabasca University June 19, 2009

2. The Procedural Reasoning System Beliefs Plan library Sensor input Interpreter Action output Desires Intentions Athabasca University

3. Introduction • Jason is a fully-fledged interpreter for an extended version of AgentSpeak, a BDI agent-oriented logic programming language, and is implemented in Java. Using SACI or JADE, a multi-agent system can be distributed over a network effortlessly. Athabasca University

4. Saci (Simple Agent Communication Infrastructure) • is a tool that facilitates the task of programming communication among distributed agents. • The main features provided by Saci are an API to manipulate KQML messages like composing, sending and receiving and tools to provide useful services in a distributed environment like • agent name service, • directory service, • remote launching, • communication debug, to name a few. Athabasca University

5. MAS Configuration File • Simple way of defining a multi-agent system MAS my_system { infrastructure: Jade environment: MyEnv ExecuctionControl: … agents: ag1; ag2; ag3; } Athabasca University

6. MAS Definition (Cont.) • Infrastructure options: Centralized, Saci, Jade • Easy to define the host where agents and the environment will run • If the file name with the code is unusual agents: ag1 at host1.dur.ac.uk; agents: ag1 file1; Athabasca University

7. When should I Use the SACI or JADE Infrastructure? • The centralized infrastructure does not support: • Execution of the agents at distributed hosts, and • Interoperability with non-Jason agents • If you need any of these features, you should choose the SACI or JADE infrastructures or implement/plug a new infrastructure for/into Jason yourself). Athabasca University

8. Execution Modes 3 3 2 2 1 1 Asynchronous Synchronous Debug Athabasca University

9. Jason Reasoning Cycle Athabasca University

10. Jason Reasoning Cycle Athabasca University

11. Jason Reasoning Cycle Athabasca University

12. Belief Annotations • Annotated predicate: ps(t1, …, tn) [a1, …, am] where ai are first order terms • All predicates in the belief base have a special annotation source(si) where si belongs to {self, percept} or AgId Athabasca University

13. Example of Annotations • An agent’s belief base with a user-defined doc annotation (degree of certainty --- doc) blue(box1)[source(ag1)] red(box1)[source(percept)] colourblind(ag1)[source(self), doc(0.7)] lier(ag1)[source(self), doc(0.2)] Athabasca University

14. Plan Annotations • Plan labels also can have annotations (e.g. to specify meta-level information) • Selection functions (Java) can use such information in plan/intention selection • Possible to change those annotations dynamically (e.g., to update priorities) • Annotations go in the plan label Athabasca University

15. Annotated Plan Example @aPlan[ chance_of_success(0.3), usual_payoff(0.9), any_other_property] +!g(X) : c(t) <- a(X). Athabasca University

16. Handling Plan Failure • Goal-deletion events were syntactically defined, but no semantics • We use them for a plan failure handling mechanism (probably not what they were meant for) • Handling plan failure is very important as agents are situated in dynamic environments • A form of “contingency (机动) plan”, possibly to “clean up” before attempting another plan Athabasca University

17. Contingency Plan Example • To create an agent that is blindly committed to goal g: +!g : g <- true +!g : … <- … ?g … -!g : true <- !g Athabasca University

18. Internal Actions • Unlike actions, internal actions do not change the environment • Code to be executed as part of the agent reasoning cycle • AgentSpeak is meant as a high-level language for the agent’s practical reasoning • Internal actions can be used for invoking legacy code elegantly Athabasca University

19. Internal Actions (Cont.) • Libraries of user-defined internal actions lib_name.action_name(...) • Pre-defined internal actions have an empty library name • Internal action for communication .send(r,ilf,pc) where ilf ∈ {tell, untell, achieve, unachieve, askOne, askAll, askHow, tellHow, untellHow} Athabasca University

20. Internal Actions (Cont.) • Examples of BDI-related internal actions: .desire(literal) .intend(literal) .drop_desires(literal) .drop_intentions(literal) Many others available for: printing, list/string operations, manipulating the beliefs/annotations/plan library, creating agents, waiting/generation events, etc. Athabasca University

21. A Jason Plan +green_patch(Rock) : ~battery_charge(low) & .desire(at(_)) <- .drop_desires(at(_)); dip.get_coords(Rock, Coords); !at(Coords); !examine(Rock). Athabasca University

22. AgentSpeak X Prolog • With the Jason extensions, nice separation of theoretical and practical reasoning • BDI architecture allows • long-term goals (goal-based behavior) • reacting to changes in a dynamic environment • handling multiple foci of attention (concurrency) • Acting on an environment and a higher-level conception of a distributed system • Direct integration with Java Athabasca University

23. Agent Customization • Users can customize the Agent class to define the selection functions, social relations for communication, and belief update and revision (buf and brf) • selectMessage() • selectEvent() • selectOption() • selectIntention() • socAcc() • buf() • brf() Athabasca University

24. Overall Agent Architecture • Users customize the AgentArch class to change the way the agent interacts with the infrastructure: perception, action, and communication • Helps switching between simulation for testing and real deployment • perceive() • act() • sendMsg() • broadcast() • checkMail() Athabasca University

25. Belief Base Customization • Logical belief base might not be appropriate for large applications • Jason has an alternative belief base combined with a database • Users can create other customization • add() • remove() • contain() • getRelevant() Athabasca University

26. Customized MAS MAS Custom { agents: a1 agentClass MyAg agentBaseClass MyAgArch beliefBaseClass Jason.bb.JDBCPersistentBB( “org.hsqldb.jdbcDriver”, “jdbc:hsqldb:bookstore”, … ”[count_exec(1, tablece)]”); } Athabasca University

27. An Example • Jason is available Open Source under GNU LGPL at: http://jason.sf.net Athabasca University

28. An Agent-based Auction Game (Simplified)Goal Overview Diagram Athabasca University

29. System Overview Diagram Athabasca University

30. auction.mas2j // creates an MAS called auction MAS auction { infrastructure: Centralised agents: ag1; ag2; ag3; auctioneer agentArchClass AuctioneerGUI; } Athabasca University

31. Agent overview diagram:Agent Auctioneer Athabasca University

32. auctioneer.asl // this agent starts the auction and identify the winner /* beliefs and rules */ all_bids_received(N) :- .count(place_bid(N,_),3). /* plans */ +!start_auction(N) : true // this goal is created by the GUI of the agent <- -+auction(N); -+winner(N, noone, 0); .broadcast(tell, auction(N)). // receive bid and check for new winner @pb1[atomic] +place_bid(N,V)[source(S)] : auction(N) & winner(N,CurWin,CurVl) & V > CurVl <- -winner(N,CurWin,CurVl); +winner(N,S,V); .print("New winner is ",S, " with value ",V); !check_end(N). @pb2[atomic] +place_bid(N,_) : true <- !check_end(N). +!check_end(N) : all_bids_received(N) & winner(N,W,Vl) <- .print("Winner is ",W," with ", Vl); show_winner(N,W); // show it in the GUI .broadcast(tell, winner(W)); .abolish(place_bid(N,_)). +!check_end(_). Athabasca University

33. Agent overview diagram: Agent ag1 Athabasca University

34. ag1.asl // this agent always bids 6 +auction(N)[source(S)] : true <- .send(S, tell, place_bid(N,6)). Athabasca University

35. Agent overview diagram: Agent ag2 Athabasca University

36. ag2.asl // This agent usually bids 4, when it has an alliance with ag3, it bids 0 default_bid_value(4). ally(ag3). +auction(N)[source(S)] : not alliance <- ?default_bid_value(B); .send(S, tell, place_bid(N,B)). +auction(N)[source(S)] : alliance <- .send(S, tell, place_bid(N,0)). // alliance proposal from another agent +alliance[source(A)] : .my_name(I) & ally(A) <- .print("Alliance proposed by ", A); ?default_bid_value(B); .send(A,tell,bid(I,B)); .send(A,tell,alliance(A,I)). Athabasca University

37. Agent overview diagram: Agent ag3 Athabasca University

38. ag3.asl // this agent bids 3, if it looses 3 auctions, it proposes an alliance to ag2 and therefore it bids 7 (3 from itself + 4 from ag2) default_bid_value(3). ally(ag2). threshold(3). +auction(N)[source(S)] : (threshold(T) & N < T) | (.my_name(I) & winner(I) & ally(A) & not alliance(I,A)) <- !bid_normally(S,N). +auction(N)[source(S)] : .my_name(I) & not winner(I) & ally(A) & not alliance(I,A) <- !alliance(A); !bid_normally(S,N). @palliance +auction(N)[source(S)] : alliance(_,A) <- ?default_bid_value(B); ?bid(A,C); .send(S, tell, place_bid(N,B+C)). +!bid_normally(S,N) : true <- ?default_bid_value(B); .send(S, tell, place_bid(N,B)). @prop_alliance[breakpoint] +!alliance(A) : true <- .send(A,tell,alliance). Athabasca University

39. Protocol auction Athabasca University

40. Protocol alliance Athabasca University

41. console [auctioneer] New winner is ag1 with value 6 [auctioneer] Winner is ag1 with 6 [auctioneer] New winner is ag3 with value 3 [auctioneer] New winner is ag1 with value 6 [auctioneer] Winner is ag1 with 6 [auctioneer] New winner is ag2 with value 4 [ag2] Alliance proposed by ag3 [auctioneer] New winner is ag1 with value 6 [auctioneer] Winner is ag1 with 6 [auctioneer] New winner is ag1 with value 6 [auctioneer] New winner is ag3 with value 7 [auctioneer] Winner is ag3 with 7 Athabasca University

42. Athabasca University