Multiagent Planning with Factored MDPs
This talk presents a framework for approximate planning that can exploit structure in a factored MDP to solve problems with many trillions of states and actions. The talk will focus on three key elements:
- Factored Linear Programs -- A novel LP decomposition technique, using ideas from inference in Bayesian networks, which can yield exponential reductions in planning time.
- Distributed Coordination -- A distributed multiagent decision making algorithm, where the coordination structure arises naturally from the system dynamics.
- Generalization in Relational MDPs -- A method for learning general solutions from solved tasks, that allows us to act in new scenarios without replanning.
We demonstrate our approach on the task of multiagent coordination in a real strategic computer war game.