Multi-Agent Planning

1
Multi-Agent Planning

2
Complexity of Negotiation Tasks Abstraction
Analysis
Tasks as Agents negotiating for resources
Self-interested Selfish Agents
I need resource X
3
Complexity of Negotiation Tasks Abstraction
Analysis
Tasks as Agents negotiating for resources
Self-interested Collaborating Agents
Self-interested Selfish Agents
I need resource X
I need resource X, and it will enable other
agents to use it later
4
Complexity of Negotiation Tasks Abstraction
Analysis
Tasks as Agents negotiating for resources
Global Welfare Oriented Collaborating Agents
Self-interested Collaborating Agents
Self-interested Selfish Agents
I need resource X
I need resource X, and it will enable other
agents to use it later
I need resource X, and it will allow us to
improve our joint success
5
Complexity of Negotiation Tasks Abstraction
Analysis
Tasks as Agents negotiating for resources
Global Welfare Oriented Collaborating Agents
Self-interested Collaborating Agents
Self-interested Selfish Agents
Distributed Constraint Satisfaction
Distributed Planning with Conjunctive Goals
Distributed (Hierarchical) Planning with
Disjunctive Goals
Technology Reusage
6
Formal Modeling Complexity Analysis More and
more important!
Distributed (Hierarchical) Planning with
Disjunctive Goals
Distributed Planning with Conjunctive Goals
Distributed Constraint Satisfaction

• As the negotiation systems are getting more
  complex, we need more and more advanced
• Formal problem modeling complexity analysis,
• Structural analysis, and
• Development of scalable generic negotiation
  protocols

7
Exploiting Structure is Crucial!
Scheduling
In terms of worst-case computational complexity,
mixed scheduling/planning is significantly harder
than scheduling. Therefore, need to exploit
problem structure to tame computational
complexity.
8
Example (Nasa)
Planning is hard find right sequence of actions
10 actions, 10! 3 x 106
Contingency planning is really hard 10224
possible plans!
10 x 92 x 84 x 78 x x 2256