1 / 68

Dr. Muhammad Adnan Hashmi

An Agent Oriented Programming Language integrating Temporal Planning and the Plan Coordination Mechanisms. Dr. Muhammad Adnan Hashmi. Outline. Introduction Background Problem Overview Plan Coordination Mechanisms Coordinated Planning Problem Proactive-Reactive Coordination Problem

felice
Download Presentation

Dr. Muhammad Adnan Hashmi

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. An Agent Oriented Programming Language integrating Temporal Planning and the Plan Coordination Mechanisms Dr. Muhammad Adnan Hashmi

  2. Outline • Introduction • Background • Problem Overview • Plan Coordination Mechanisms • Coordinated Planning Problem • Proactive-Reactive Coordination Problem • P-CLAIM: AOP Language supporting Temporal Planning • Language Definition • Planning Mechanism • Conclusion

  3. Outline • Introduction • Background • Problem Overview • Plan Coordination Mechanisms • Coordinated Planning Problem • Proactive-Reactive Coordination Problem • P-CLAIM: AOP Language supporting Temporal Planning • Language Definition • Planning Mechanism • Conclusion

  4. What is an Agent? • An agent is a computer system capable of autonomous action in some environment in order to meet its design objectives (goals). • Achievement of goals is important for agents

  5. Planning • Finding a sequence of actions that takes agent from Initial State to Goal State Initial state:clear(c),on(c,a),ontable(a),clear(b),ontable(b),handempty Goal:on(a,b), on(b,c) a c b a b c a c b b a b a b c a c c

  6. Multi-Agent Planning • Multiple agents sharing the same environment • Independent planning is not sufficient • Coordination of plans is vital • Coordination of plans • Removing conflicts (negative interactions) • Utilizing help relations (positive interactions) DA DA A A DB DB B B

  7. How to Program Agents? • Agent Oriented Programming (AOP) Languages allow to program intelligent and autonomous agents • Main Characteristics • Mental State: Beliefs, Goals, Commitments • Reasoning Mechanism • Capabilities, Services • Communication • Some Languages • Agent-0 [Shoham 1993], 2APL [Dastani 2008], AgentSpeak (L) [Rao 1996]

  8. Outline • Introduction • Background • Problem Overview • Plan Coordination Mechanisms • Coordinated Planning Problem • Proactive-Reactive Coordination Problem • P-CLAIM: AOP Language supporting Temporal Planning • Language Definition • Planning Mechanism • Conclusion

  9. Problem Statement • Current AOP languages • Follow a reactive (PRS based) approach • Do not support temporal planning • Only a few support planning • Problems • Execution without planning may result in the goal failures • Agent can reach a dead end • Conflicts can arise among different agents • Actions’ duration is important • Real world actions take place over a timespan

  10. Objectives • Propose a programming language that endows agents with planning skills • Has temporal planning • Deals with uncertainty of the environment • Incorporate reactivity by dealing with on the fly goals having different priorities • Propose coordination mechanisms for the plans having different priorities

  11. Outline • Introduction • Background • Problem Overview • Plan Coordination Mechanisms • Coordinated Planning Problem • Proactive-Reactive Coordination Problem • P-CLAIM: AOP Language supporting Temporal Planning • Language Definition • Planning Mechanism • Conclusion

  12. Assumptions • Two agents α and β sharing the same environment • Agent α having higher priority (reactive) goals • Agent β having normal priority (proactive) goals • Actions have: • Preconditions • Add effects • Delete effects • Two possible conflicts among plans • Causal link threat • Parallel actions interference

  13. Two Possible Conflicts • Causal Link (A1, A2, p) • Action A1 adds an effect p • Action A2 needs this effect • No action between A1 and A2 adding p • Causal Link Threat • If an action A deletes p and lies between A1 and A2, then A threatens the causal link (A1, A2, p) p A1 A2 Threat A ¬p

  14. Two Possible Conflicts • Causal Link (A1, A2, p) • Action A1 adds an effect p • Action A2 needs this effect • No action between A1 and A2 adding p • Parallel Actions Interference • Actions A1 and A2 lie in parallel • Either one of them deletes the preconditions or add effects of the other A1 p A1 ¬p p A2 ¬p A2

  15. Outline • Introduction • Background • Problem Overview • Plan Coordination Mechanisms • Coordinated Planning Problem • Proactive-Reactive Coordination Problem • P-CLAIM: AOP Language supporting Temporal Planning • Language Definition • Planning Mechanism • Conclusion

  16. Coordinated Planning Problem • Prerequisite: • Plan Pα of Agent α • Our Aim: • Compute a Plan Pβ for Agent β • Has no conflict with Pα • Avails the cooperative opportunities offered by Pα • Solution: • Non Temporal Domains  µ-SATPLAN • Temporal Domains  Coordinated-Sapa

  17. SATPLAN [Kautz and Selman 2006] A Classical Planner that Finds Optimal Plans in Non-Temporal Domains

  18. SATPLAN We Use GraphPlan Encoding Propositional formula in conjunctive normal form (CNF) • Planning Problem • Init State • Goal • Actions Compiler (encoding) Simplifier (polynomial inference) CNF Increment plan length If unsatisfiable CNF satisfying model Decoder Solver (SAT engine/s) Plan

  19. Constructing the planning graph • Level P1: All literals from the initial state • Add an action in level Ai if all its preconditions are present in level Pi • Add a proposition in level Pi if it is the effect of some action in level Ai-1 • Maintain a set of exclusion relations to eliminate incompatible propositions and actions

  20. GraphPlan Encoding • Can create such constraints for every node in the planning graph • Only involves facts and actions in the graph Pre1 Act1 Fact Pre2 Act2 Fact  Act1  Act2 Act1  Pre1  Pre2 ¬Act1  ¬Act2

  21. µ-SATPLAN An Extension of SATPLAN that Computes Coordinated Plans in Non-Temporal Domains

  22. Handling Causal Link Threat • While constructing the planning graph for Agent β, don’t add an action O at level Ai if • It has an effect ¬p, and • There is a causal link (Aj, Ak, p) in plan Pα, and • j ≤ i ≤ k Action O threatens Causal Link (Aj, Ak, p)

  23. Handling Positive Interactions & Parallel Actions Interference • For each time step i in the plan of Agent α, create an action NoName(i) • Add(NoName(i))  All the effects added by Pαat time step i • Del(NoName(i))  All the effects deleted by Pαat time step i • Pre(NoName(i))  All the preconditions of actions of Pαat time step i • Explicitly add all the NoName actions in the planning graph of Agent β

  24. Level 0 Level 1 Level 2 a0 a0 a0 NoName(0) NoName(1) a2 a2 a1 a1 a1 β2 β4 a8 a4 a4 a4 a4 β1 a5 a5 a5 β3 a7 a7 a6 a6 a6 a6 Handling Positive Interactions • Pα = {α1(0), α2(0), α3(1)} • Eff(α1) = a0, Eff(α2) = a1, Eff(α3) = a2 • Eff( NoName(0) ) = {a0,a1}, Eff( NoName(1) ) = {a2} Planning Graph of Agent β

  25. Level 0 Level 1 Level 2 a0 a0 a0 NoName(0) NoName(1) a2 a2 a1 a1 a1 β2 β4 a8 a4 a4 a4 a4 β1 a5 a5 a5 β3 a7 a7 a6 a6 a6 a6 Handling Positive Interactions • Add NoName actions as unary clauses to CNF • NoName(0) • NoName(1) • Partial CNF Sentence • a8  β4 • β4  a0  a7 • a0  β2  NoName(0) • a7  β3 • β2  a5 • … Solution Problem

  26. Example Plan Generated (Logistics) Pα Pβ

  27. Outline • Introduction • Background • Problem Overview • Plan Coordination Mechanisms • Coordinated Planning Problem • Proactive-Reactive Coordination Problem • P-CLAIM: AOP Language supporting Temporal Planning • Language Definition • Planning Mechanism • Conclusion

  28. Proactive-Reactive Coordination • Prerequisite: • Reactive plan Pα of Agent α • Proactive plan Pβ of Agent β • Our Aim: • Modify plan Pβ such that: • It has no conflict with Pα • Avails the cooperative opportunities offered by Pα • Solution: • Plan Merging Algorithm

  29. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 A B C D Case Study Rescue Agent : α Analyzer Agent : β • Tasks of Rescue Agent • Rescue the Victims • Tasks of Analyzer Agent • Analyze the goal cells • Call the central agent • Constraints • One agent in a cell • Hyper energy cells • Needs fuel or energy to enter • Agent should have key to open door

  30. Conflict Resolution • Threat-Repair Link (A1, A2, p) • Action A1 deletes p • A2 is a subsequent action and adds p • A1 is called Threat Action • A2 is called Repair Action p B1 B2 Threat A1 -p A2 p Repair

  31. Valid and Possibly Valid Time Stamps • Possibly Valid Time Slot for an action A • All preconditions are met • No parallel actions interference P[1]  a b h P[2]  b c -d P[3]  c e P[4]  e f P[5]  f g P[6]  g i P[7] • i • h g P[1]  b d -h • Valid Time Slot for an action A • All preconditions are met • No parallel actions interference • Either: • No causal link threat • Repair Action exist before the deadline P[1]  a b h P[2]  b c -d P[3]  c e P[4]  e f P[5]  f g P[6]  g i P[7] • i • h g P[2] k P[1]  b d -h  d P[3] • k h

  32. Plan Merging Algorithm • Fix all the actions of Reactive Plan Pα on timeline • For every action CA of Proactive Plan • Search for the first Possibly Valid Time Slot T on timeline • Reason about the time slot T • There could be 5 cases at T

  33. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 A B C D Plan Merging Algorithm Case 1: No causal link threat by CA at T • Assign Time Slot T to CA EXAMPLE • Current Action: Move(A1, A2) • Returned Time Slot: 0 - 5 • Any Threat? : No • Assign Time Slot 0 – 5 to CA

  34. Plan Merging Algorithm Case 2: CA threatens a Causal Link but Repair Action exist • Assign Time Slot T to CA • Save a Possible Threat <ThreatAction, RepairAction, Deadline> EXAMPLE • Current Action: Move(A4, A5) • Time Slot: 20 - 25 • Any Threat? : Yes (Agent α needs A5 at time 40-45) • Repair Action: Move(A5, A6) • Assign Time Slot 20 - 25 to Move (A4, A5) • Save <Move(A4, A5), Move(A5, A6), 40> 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 A B C D

  35. Plan Merging Algorithm Case 3: It is a Repair Action but can not meet a deadline of some Threat Action • Backtrack to the Threat Action,find another time stamp EXAMPLE • Current Action: Move (A8, A9) • Returned Time Slot: 50 - 55 • Any Threat?: Yes (Agent α needs A9 at 85-110) • Repair Action : Move (A9, B9) • Save <Move(A8, A9), Move(A9, B9), 85> • Next Action: AnalyzeCell (A9) • Time Slot Assigned: 55 - 70 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 A B C D

  36. Plan Merging Algorithm Attention • Next Action: CallCentral (A9) • Time Slot Assigned: 80 – 90 • Next Action: Move (A9, B9) • Is it a Repair Action? : Yes • Meet all deadlines?: No (Agent α needs A9 at 85) • Backtrack to action Move(A8, A9) • Find another Time Slot • New Time Slot: 110 – 115 (Valid Time Slot) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 A B C D

  37. Plan Merging Algorithm Case 4: All the effects of CA are already achieved WHAT TO DO? • Mark CA as redundant POST PROCESSING • Remove all redundant actions from the plan • Recursively remove all actions which achieve only the preconditions of removed action

  38. Plan Merging Algorithm EXAMPLE • Current Action: OpenDoor (C11) • Returned Time Slot: 172 - 175 • Redundant(OpenDoor(C11))  true • BecauseopenedDoor(C11) is true at time 172 • When the plan is returned • Remove OpenDoor(C11) from plan • Also remove TakeKey(C11, key1) from plan 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 A B C D

  39. Plan Merging Algorithm Case 5: Action CA’s preconditions can not be achieved • Remove action CA from the plan and compute a plan to achieve effects of CA • I = State just before CA • G = Effects (CA) • Plan should have no conflict with Reactive Plan Pβand if CA is a repair action, repair effects must meet their deadline • ReplacementPlan = Coordinated-Sapa (I, G, Pβ) • If a plan is returned, replace the removed actions with the plan • If a deadline is violated, backtrack to the threat action • If no plan possible, then remove another action CA + 1 • G = G U Effects (CA + 1) \ Pre (CA + 1) Use Coordinated-Sapa

  40. Plan Merging Algorithm EXAMPLE • Current Action: TakeEnergy(B13, energy1) • Preconditions can not be achieved • Repair the plan 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 A B C D

  41. Plan Repair Algorithm • Create a CPP by removing TakeEnergy(B13, energy1) • I = { at(β, B13), at(energy1, B13), at(energy2, B13) } • G = { hasEnergy(β, energy), at(β, B13)} • Call Coordinated-Sapa to solve this CPP • Coordinated-Sapa returns fail • Why? energy2 is also needed by Agent α 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 A B C D

  42. Plan Repair Algorithm • Create another CPP by removing Move(B13, A12) • I = { at(β, B13), at(energy1, B13), at(energy2, C15) } • G = { at(β, A12) } • Call Coordinated-Sapa to solve this CPP • A plan is returned to enter A12 by taking the fuel from D14 POST PROCESSING • This plan will become a replacement for both TakeEnergy(B13, energy1) and Move(B13, A12) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 A B C D

  43. Outline • Introduction • Background • Problem Overview • Plan Coordination Mechanisms • Coordinated Planning Problem • Proactive-Reactive Coordination Problem • P-CLAIM: AOP Language supporting Temporal Planning • Language Definition • Planning Mechanism • Conclusion

  44. P-CLAIM • An AOP language having: • Cognitive aspects specific to intelligent agents • Communication primitives • Mobility primitives • Temporal planning capability • P-CLAIM Agent: • Is autonomous, intelligent and mobile • Has a mental state containing knowledge, goals, and capabilities • Is able to communicate with other agents • Entails a planning based behaviour • Achieves goals based on their priorities • Maintains the stability of the plan in the dynamic environments

  45. Defining P-CLAIM Agent defineAgent agentName{ parent = null | agentName ; knowledge = null ;| {knowledge1; …; knowledgem} goals = null;| {goal1; … ; goaln} activities = null ; {activity1 … activityo} actions = {action1 … actionp} agents = null ; | {agName1, agName2, …, agNameq} }

  46. Activities defineAgent agentName{ parent = null | agentName ; knowledge = null ;| {knowledge1; …; knowledgem} goals = null;| {goal1; … ; goaln} activities = null ; {activity1 … activityo} actions = {action1 … actionp} agents = null ; | {agName1, agName2, …, agNameq} } activityity = name { message = null | message ; conditions = null | condition ; do { process } effects = null ;| { effect1 ; …; effectf } }

  47. Actions defineAgent agentName{ parent = null | agentName ; knowledge = null ;| {knowledge1; …; knowledgem} goals = null;| {goal1; … ; goaln} activities = null ; {activity1 … activityo} actions = {action1 … actionp} agents = null ; | {agName1, agName2, …, agNameq} } action = name { message = null | message ; conditions = null | condition ; do { process } duration = dur ; }

  48. Outline • Introduction • Background • Problem Overview • Plan Coordination Mechanisms • Coordinated Planning Problem • Proactive-Reactive Coordination Problem • P-CLAIM: AOP Language supporting Temporal Planning • Language Definition • Planning Mechanism • Conclusion

  49. Agent Definition to Planning (Translator) Agent Description File Knowledge Goals Activities Actions Translator (JavaCC) Initial State Goals Methods Operators Problem File Domain File Planner

  50. Agent Life Cycle

More Related