Intelligent Agents: Technology and Applications Coordination and Planning

1
Intelligent Agents Technology and
ApplicationsCoordination and Planning

• IST 597B
• Spring 2003
• John Yen

2
Learning Objectives

• Given an agent coordination and planning problem,
  be able to identify potential conflicts.

• Given a potential conflict, be able to design
  solutions/algorithms for resolving the conflict.

3
An Example of Single Agent Planning

• Initial State On(C A) and OnTable(B)

C
A
B

• Goal State On(A B) and On(B C)

A
B
C
4
Operators

• PutOnTable(X) Put X on the table
• Precondition Clear(X)
• Effect OnTable(X)
• PutOn(X, Y) Put X on Y.
• Precondition Clear(X) and Clear(Y)
• Effect On(X, Y), Not Clear(Y)

5
Linear (STRIPS) Planner

• Initialize the goal list with all conditions in
  the goal state
• Repeat
• Select a goal from the goal list
• Find an operator whose effect matches the goal,
• Add the operator to the front
  of the current plan (a sequence
  of actions)
• For each precondition of the operator,
• If it is not satisfied by the
  initial state, add it to the goal list.
• Until the goal list is empty

6
Problem 1 (5 points)

• What problems are encountered in applying STRIPS
  planning to the problem? What causes the
  problem?

7
The Sussman Anomaly

• Introduces the issue of goal interaction (i.e.,
  conflicts between actions in a plan) by Gerald
  Jay Sussman (1974)
• Reveals the problem of linearity assumption of
  STRIPS planners
• Motivates the second generation planners
  (nonlinear planners)

8
Initial State
Clear (C), On (C,A) On Table(A) OnTable(B) Clear
(B)
threat
Goal State
On(B,C) On(A,B)
OnTable(C) Clear(A) Clear(B)
Clear(C)
9
Conflicts in Plans

• A planned action a1 by agent A creates a conflict
  if
• It prevents agent B to perform a planned action
  a2 (i.e., precondition conflict), or
• It interferes with the desired conditions in the
  goal state (i.e., goal conflict)

10
Detecting Conflicts

• Detecting a potential violation of a constraint
  for protecting
• a precondition of an action or
• a condition in the goal state.

11
Target of Protection Causal Dependency

• A causal dependency exists between an action A1
  and action A2 if the effect of A1 establishes a
  precondition of A2.
• A causal dependency exists between an action A1
  and the goal state S if the effect of A1 achieves
  a goal condition in S.

12
Initial State
Clear (C), On (C,A) On Table(A) OnTable(B) Clear
(B)
threat
Goal State
On(B,C) On(A,B)
OnTable(C) Clear(A) Clear(B)
Clear(C)
13
Solution for Handling Threats to Causal Dependency

• Promotion Move the action that creates the
  threat BEFORE the threatened causal dependency.
• Demotion Move the action that creates the
  threat AFTER the threatened causal dependency.

14
Initial State
Clear (C), On (C,A) On Table(A) OnTable(B) Clear
(B)
Threat resolved
before
Goal State
On(B,C) On(A,B)
OnTable(C) Clear(A) Clear(B)
Clear(C)
15
Problem 2 (5 points) Consider the problem
described in p. 428 and Fig. 8.24 of the text.

• Is this conflict about resource the same as those
  revealed in the Sussman Anomaly? Explain your
  answer.
• Does the human solution described in the book
  suggest additional conflict resolution schemes
  (beyond promotion and demotion) in the case of
  multi-agent planning? If so, describe these
  additional conflict resolution schemes in
  multi-agent planning.

16
Solutions for Handling Threats in Multi-agent
Planning

• Promotion
• Demotion
• Consider alternative ways to perform a task
• Replace actors through
• Negotiations
• Delegations
• Bidding