An on-line algorithm for dynamic reinforcement learning and planning in reactive environments

An on line learning algorithm for reinforcement learning with continually running recur rent networks in non stationary reactive environments is described Various kinds of rein forcement are considered as special types of input to an agent living in the environment The agent s only goal is to maximize the amount of reinforcement received over time Supervised learning techniques for recurrent networks serve to construct a di erentiable model of the environmental dynamics which includes a model of future reinforcement This model is used for learning goal directed behavior in an on line fashion The method extends work done by Munro Robinson and Fallside Werbos Widrow and Jordan The possibility of using the system for planning future action sequences is investigated and this approach is compared to approaches based on temporal di erence methods A connection to meta learning learning how to learn is noted