next up previous
Next: Recurrent Q-learning Up: Partially Observable Environments Previous: State-Free Stochastic Policies

Policies with Internal State

The only way to behave truly effectively in a wide-range of environments is to use memory of previous actions and observations to disambiguate the current state. There are a variety of approaches to learning policies with internal state.

Leslie Pack Kaelbling
Wed May 1 13:19:13 EDT 1996