1. Feb 24, 2003 Agent-based Proactive Teamwork John Yen University Professor of IST School of Information Sciences and Technology The Pennsylvania State University

2. Feb 24, 2003 Motivation Psychological Studies about Effective Human Teamwork Indicated that Team members can anticipate needs of team mates Team members can anticipate needs of team mates Team members can offer relevant information proactively. Team members can offer relevant information proactively. These teamwork behaviors are based on an overlapping shared mental model. These teamwork behaviors are based on an overlapping shared mental model.

3. Feb 24, 2003 A Response to the Challenges Empower agents with ``shared mental models" to enable them to better Empower agents with ``shared mental models" to enable them to better Anticipate information needs of teammates in a dynamic environment Anticipate information needs of teammates in a dynamic environment Offer relevant information proactively to soldiers Offer relevant information proactively to soldiers

4. Feb 24, 2003 Shared Mental Model Shared Ontology Shared Ontology Shared Goals Shared Goals Shared Team Structure Shared Team Structure Shared Team Collaboration Process Shared Team Collaboration Process Shared Belief about the Team Shared Belief about the Team Shared Belief about the World Shared Belief about the World Shared Hypotheses about the Enemy Shared Hypotheses about the Enemy

5. Feb 24, 2003 Major Features of CAST Use a high-level language to describe teamwork knowledge Use a high-level language to describe teamwork knowledge Capture “shared mental model” about team structure and process Capture “shared mental model” about team structure and process Infers information needs (from SMM) Infers information needs (from SMM) induces proactive information exchanges induces proactive information exchanges

6. Feb 24, 2003 Anticipating Information Needs of Teammates Team Plan Responsibilities of Tasks Preconditions of Tasks Information Needs Dynamic Task Allocation Who needs what

7. Feb 24, 2003 Dynamic Task Allocation Team Plan Constraints for Task Allocation Roles of Agents in the Team My Belief about The World My Belief about Teammates Dynamic Task Allocation

8. Feb 24, 2003 Proactive Information Delivery My Belief about Teammates Information Needs Information Match ? Communication Strategy Does he/she know ? How to inform him/her?

9. Feb 24, 2003 CAST Agent Architecture Team Knowledge (MALLET) Responsibilities (Petri Nets) Belief Domain Knowledge Responsibility Selection Identify Info Needs Information Needs Information Belief Update Act on Info Needs

10. Feb 24, 2003 Shared Mental Model in CAST Shared Mental Model in CAST Prolog knowledge base: belief Prolog knowledge base: belief MALLET: High-level language for representing team knowledge MALLET: High-level language for representing team knowledge Petri Nets: An agent’s internal representation of the dynamic teamwork processes and related information requirements Petri Nets: An agent’s internal representation of the dynamic teamwork processes and related information requirements

11. Feb 24, 2003 Relationships between SMM Components MALLET Knowledge Base Prolog Knowledge Base (belief) MALLET Compiler Petri Net (team process) query reply CAST Kernel

12. Feb 24, 2003 (team-plan T1 () (process (par (kill-wumpuses) (collect- gold)))) (process (par (kill-wumpuses) (collect- gold)))) (team-plan kill-wumpuses () (agent-bind ?s (play-role ?s scout)) (agent-bind ?s (play-role ?s scout)) (agent-bind ?f (play-role ?f fighter)(closest ?f wumpus)) (agent-bind ?f (play-role ?f fighter)(closest ?f wumpus)) (process (while ((wumpus ?x) (not (dead ?x)))) (process (while ((wumpus ?x) (not (dead ?x)))) (seq (do ?s (find-wumpus ?x)) (seq (do ?s (find-wumpus ?x)) (do ?f (move-to-wumpus ?x)) (do ?f (move-to-wumpus ?x)) (do ?f (shoot-wumpus ?x))))) (do ?f (shoot-wumpus ?x))))) (team-plan find-gold () (agent-bind ?c (play-role ?c carrier)) (agent-bind ?c (play-role ?c carrier)) (process (while (true) (if (see ?any-agent glitter) (process (while (true) (if (see ?any-agent glitter) (do ?c (carrier-pickup gold)))) (do ?c (carrier-pickup gold)))) start findshootmove wumpus exists glitter done no wumpuses left pickup

13. Feb 24, 2003 Domain Simulator Team Process ExecutionTeam Process Specification CAST Software Architecture MALLET Team Structure Process Net Knowledge Base Kernel Agent Selection Goal Selection Domain World Communication Fact Base & Domain Policy Monitor Inference Engine Parser Domain Adapter Inference Engine

14. Feb 24, 2003 CAST Development Environ. Circles are places and hold tokens denoting current execution state. Red indicates the presence of a token. Rectangles are transitions and are tested and executed when preceding places have tokens.

15. Feb 24, 2003 Theoretical Foundations of CAST What are the principles that derive “information needs”? What are the principles that derive “information needs”? What is the semantics of proactive information delivery? What is the semantics of proactive information delivery? How does an agent choose among different proactive communication strategies? How does an agent choose among different proactive communication strategies?

16. Feb 24, 2003 Agent Theories Joint Intention theory (Cohen & Levesque) Joint Intention theory (Cohen & Levesque) SharedPlan Theory (Grosz & Kraus) SharedPlan Theory (Grosz & Kraus)

17. Feb 24, 2003 Two Types of Information needs Action-performing information needs Action-performing information needs enables an agent to perform certain (complex) actions, which contributes to an agent's individual commitments to the whole team. enables an agent to perform certain (complex) actions, which contributes to an agent's individual commitments to the whole team. Goal-protection information needs Goal-protection information needs allows an agent to protect a goal from potential threats that may result in a conflict with the goal. allows an agent to protect a goal from potential threats that may result in a conflict with the goal.

18. Feb 24, 2003 Semantics of Assert Performative (Cohen & Levesque) The speaker attempts to establish a joint belief with the receiver that The speaker attempts to establish a joint belief with the receiver that The speaker believes the information The speaker believes the information Whether the receives accept the belief depends on the reply (accept or reject) Whether the receives accept the belief depends on the reply (accept or reject)

19. Feb 24, 2003 Semantics of ProAssert The speaker attempts to establish a joint belief with the receiver that The speaker attempts to establish a joint belief with the receiver that The speaker believes the information The speaker believes the information The speaker believes that the receiver needs the information The speaker believes that the receiver needs the information

20. Feb 24, 2003 Semantics of ProAssert An attempt for the speaker (s) to establish a joint belief that An attempt for the speaker (s) to establish a joint belief that 1. S believes in the information (p) 2. S believes that the receiver (a) needs the information.

21. Feb 24, 2003 Responses to ProAssert Three possible responses to ProAssert: Accept: The receiver believes the information Accept: The receiver believes the information Reject1: The receiver refuses the information because the information is contradictory to its beliefs. Reject1: The receiver refuses the information because the information is contradictory to its beliefs.  Reject2: The receiver refuses the information because it believes it does not need the information

22. Feb 24, 2003 Inferring Proactive Assert from Information Needs Derived from the Assist Axiom in SharedPlan theory:

23. Feb 24, 2003 Choosing Communication Policies How to inform teammates? Proactive or wait for them to ask? How to inform teammates? Proactive or wait for them to ask? Factors that affect the policies Factors that affect the policies Cost of communications Cost of communications Certainty level about teammate’s needs Certainty level about teammate’s needs Consequence of communications Consequence of communications Criticality of the information to teammates Criticality of the information to teammates Load level of teammates Load level of teammates These factors depend on situations agents are in These factors depend on situations agents are in

24. Feb 24, 2003 Decision-theoretic Approach to Communication Use decision theory for choosing communication strategies Use decision theory for choosing communication strategies Assign utility to plans Assign utility to plans Assess the value of information from the utility of plans Assess the value of information from the utility of plans Specify probabilities in JARE Specify probabilities in JARE

25. Feb 24, 2003 Experiment Goal: Evaluate Two Proactive Communication Strategies Strategy Action-performing Information Needs Goal-protection Information Needs S1 Always ProInform S2 Decision-theoretic ProInform Decision-theoreticProInform

26. Feb 24, 2003 MALLET Example (action attack-enemy () (pre-cond (is-enemy_homebase ?h1) (at ?h1 ?x ?y) ) (process(par (do scoutteam seeking) (do fighterteam attacking-wumpus) (do boomerteam attacking-base ?x ?y) ) ) )

27. Feb 24, 2003 Petri Net – Process Representation Parsed from MALLET Parsed from MALLET Hierarchical structure Hierarchical structure Extended with predicate Extended with predicate Support basic processes such as sequential, parallel, and conditional Support basic processes such as sequential, parallel, and conditional

28. Feb 24, 2003 Domain Overview Two teams, attack team (blue) and defense team (red) Two teams, attack team (blue) and defense team (red) The mission of the blue team to is to destroy enemy base by flowing certain waypoints The mission of the blue team to is to destroy enemy base by flowing certain waypoints Communication risk Communication risk Different observability Different observability The red team will patrol around the waypoints and home base The red team will patrol around the waypoints and home base

29. Feb 24, 2003 Domain Simulation Interface Blue team Blue team Red team Red team Waypoint Waypoint Enemy Base Enemy Base Agent Monitor Agent Monitor World Monitor World Monitor

30. Feb 24, 2003 Demo

31. Experiment Result

32. Feb 24, 2003 Applications Training for AWACS-like synthetic task (AFOSR MURI) Training for AWACS-like synthetic task (AFOSR MURI) Support Horizontal Fusion STO (Army Research Lab) Support Horizontal Fusion STO (Army Research Lab) Negotiation among Agent Teams for Design (NSF) Negotiation among Agent Teams for Design (NSF) Supporting Analyst Teams (Lockheed Martin) Supporting Analyst Teams (Lockheed Martin)

33. Feb 24, 2003 Conclusion A computational shared mental model is critical for developing agents that support proactive teamwork. A computational shared mental model is critical for developing agents that support proactive teamwork. CAST enables proactive information delivery by anticipating needs of teammates. CAST enables proactive information delivery by anticipating needs of teammates. MALLET facilitates the reuse of teamwork knowledge MALLET facilitates the reuse of teamwork knowledge CAST achieves efficiency using shared team plans and shared policy CAST achieves efficiency using shared team plans and shared policy CAST achieves adaptability by dynamic assignment of agent responsibility CAST achieves adaptability by dynamic assignment of agent responsibility