intelligent agents: technology and applications agent teamwork ist 597b spring 2003 john yen
TRANSCRIPT
Intelligent Agents: Technology and ApplicationsAgent Teamwork
IST 597B
Spring 2003
John Yen
Learning Objectives: Given an application that involves a group of agents, be
able to identify its major characteristics (e.g., adhoc vs structured team etc).
For a specific types of agent team applications, be able to identify major issues related to the design of such agent teams.
Given an agent team applications, be able to determine whether a particular agent teamwork model/architecture (i.e., CAST) is suitable for the application.
Problem 1
Consider the following five applications involving agent teams, identify several key characteristics that are important for these applications. Use a table to compare the similarities and differences of these applications along these characteristics.
1. A team of agents that play Robot Soccer together.2. A team including robots, soldiers, and software
agents (for information fusion/delivery) in the battle field.
3. A team of software agents that support/automate the information exchanges and/or transactions of business partners (e.g., supplier of parts, manufacturers of products, distributors, ...) in a supply chain.
4. A group of agents that assist the companies they each represent to form coalitions for business opportunities.
5. A group of agents, each represent a user, interacts in an e-auction marketplace.
Characteristics
Team membership: Static vs dynamic Shared goals vs individual goals Benevolent vs selfish Hierarchical vs Egalitarian Homogenous vs Heterogeneous Level of Trust Coordination vs competition
Key Characteristics for Agent Teamwork
Benevolent (shared goals) vs selfish (individual goals) Capabilities: Homogenous vs Heterogeneous Membership of the Team: Static vs dynamic Structure of the Team: Completely predefined, partially
defined by roles, completely unspecified. Types of the structure: Hierarchical vs Egalitarian Process of the Team: Completely predetermined, partially
specified, dynamically generated. Human agent-software agent relationship: boss-assistant,
peer, trainee-coach. Relationship between members of the Group:
Cooperative, partially cooperative, competitive. The level of trust
Homework 3 (15%, team assignment, due April 8th)
Compare the similarities and differences of the five agent teams using the key characteristics identified in class. Describe a (sixth) agent team application and characterize it using the characteristics.
Problem 2: (5%) What are important issues related to these
characteristics?
Related Issues - Team Structure How to form a team? How to determine the structure of
a team?– Important if the structure of the team is dynamically determined.
How to specify roles and assign responsibilities based on roles?
– Important if the team structure is partially specified by roles. How to reconfigure a team?
– Important if members of the team may die or be overloaded How to resolve conflicts in a team?
– Important if the team does not have a hierarchical structure.
Related Issues - Team Process
How to specify, coordinate, and execute a team process?– If the process is partially/fully specified.
How to generate a team process (through planning)?– If the process is dynamically generated.
How to make sure emergent behavior achieves expected effects?– If the process is not specified.
Related Issues - Human-agent Relationship
How to give human users adequate control of agents? – If agents are assistant to human
How to enable agents to understand the mental states of human and the context of the interactions?– If agents are peer; and (to some degree) assistant.
How much can users trust agents? How to design friendly interface to enables agents
and user interact more effectively? How to make agents human-like?
Related Issues - Cooperative/Competitive
How to enable agents to negotiate with others?– If agents are partially cooperative partially
competitive
How does an agent balance intentions of others with intentions of self when they are conflicting.
– If agents are not selfish less (I.e., partially self-centered).
Related Issues - Trust
How to establish/guarantee/revise trust?– Important if trust is important
Two Conflicting Objectives of Teamwork Models Efficiency
– Higher team performance– Reduced communications
Flexibility/Adaptability– Adapt the structure of the team– Adapt the process of the team– Adapt the responsibility assignment
Goals for CAST Teamwork Model
Achieves a high-level of efficiency with a reasonable degree of adaptability for applications with role-based structure and process.
Motivation
Psychological Studies about Effective Human Teamwork Indicated that
Team members can anticipate needs of team mates Team members can offer relevant information
proactively. These teamwork behaviors are based on an
overlapping shared mental model.
Shared Mental Model
Shared Ontology Shared Goals Shared Team Structure Shared Team Collaboration Process Shared Belief about the Team Shared Belief about the World
– Shared Hypotheses about the Enemy
CAST Agent Architecture
Use a high-level language to describe teamwork knowledge
Capture “shared mental model” about team structure and process
Infers information needs (from SMM) induces proactive information exchanges
Anticipating Information Needs of Teammates
Team PlanResponsibilities
of Tasks
Preconditions ofTasks
Information Needs
DynamicTask
Allocation
Who
needs what
Dynamic Task Allocation
Team Plan
Constraints for TaskAllocation
Roles of Agents in the Team
My Belief aboutThe World
My Belief aboutTeammates
DynamicTask
Allocation
Proactive Information Delivery
My Belief aboutTeammates
Information Needs
InformationMatch ?
CommunicationStrategy
Does he/sheknow ?
How toinform him/her?
CAST Agent Architecture
TeamKnowledge (MALLET)
Responsibilities(Petri Nets)
Belief
DomainKnowledge
ResponsibilitySelection
IdentifyInfo Needs
InformationNeeds
InformationBeliefUpdate
Act onInfo Needs
Shared Mental Model in CAST Prolog knowledge base: belief MALLET: High-level language for
representing team knowledge Petri Nets: An agent’s internal
representation of the dynamic teamwork processes and related information requirements
Relationships between SMM Components
MALLETKnowledge Base
PrologKnowledge Base
(belief)
MALLETCompiler
Petri Net(team process)
query
reply
CASTKernel
(team-plan T1 () (process (par (kill-wumpuses) (collect-gold))))
(team-plan kill-wumpuses () (agent-bind ?s (play-role ?s scout)) (agent-bind ?f (play-role ?f fighter)(closest ?
f wumpus)) (process (while ((wumpus ?x) (not (dead ?x)))) (seq (do ?s (find-wumpus ?x)) (do ?f (move-to-wumpus ?x)) (do ?f (shoot-wumpus ?x)))))(team-plan find-gold () (agent-bind ?c (play-role ?c carrier)) (process (while (true) (if (see ?any-agent
glitter) (do ?c (carrier-pickup gold))))
start find shootmove
wumpus exists
glitter
done
no wumpuses left
pickup
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.
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.
Two Types of 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.
Goal-protection information needs– allows an agent to protect a goal from potential
threats that may result in a conflict with the goal.
Applications
Training for AWACS-like synthetic task (AFOSR MURI)
Support Agent-based Collaborative Mission Planning (Army Research Lab)
Simulating Digital TOC (STRICOM) Negotiation among Agent Teams for
Engineering Design (NSF)
Conclusion A computational shared mental model is critical for
developing agents that support a team involving both agents and human.
(PSU-TAMU) CAST enables proactive information delivery by anticipating needs of teammates.
MALLET facilitates the reuse of teamwork knowledge CAST achieves efficiency using shared team plans and
shared policy CAST achieves adaptability by dynamic assignment of
agent responsibility