Multiagent Epistemic Cooperation-Agnostic Planning

Abstract

PhD Defense:

In Epistemic Planning an automated system decides what goal-oriented actions to perform in a task-environment where the mental perspectives (beliefs and knowledge) of multiple agents are of prime importance. This work presents three contributions to Epistemic Planning. First, three action languages improve upon prior languages to expand the class of planning domains that can be expressed with a natural-language-like syntax. Second, a new planning algorithm generalizes prior techniques by relaxing assumptions about how states and actions are represented. This approach also incorporates the predicted behavior of independent agents (beyond the planner's control), allowing the flexibility to plan for cooperative, competitive, or neutral (self-interested) multi-agent domains. Finally, we evaluate the Multi-Agent Cooperation-Agnostic Planner, which implements these methods.

Please join meeting via Zoom:

https://tufts.zoom.us/j/96382873058

Password: see colloquium email

Dial-in not an option for this event.