Bounded Finite State Controllers

We describe a new approximation algorithm for solving partially observable MDPs. Our bounded policy iteration approach searches through the space of bounded-size, stochastic finite state controllers, combining several advantages of gradient ascent (efficiency, search through restricted controller space) and policy iteration (less vulnerability to local optima).

• Type: Conference
• Year: 2003
• Where: NIPS
• Authors: Pascal Poupart, Craig Boutilier
• Keywords: NIPS 2003, NIPS 2007, Policy Iteration, Policy Iteration Approach, Stochastic finite State