|
Abstract : |
This paper analyzes the suitability of reinforcement learning for both programming and adapting situated agents. In the the first part of the paper we discuss two specific reinforcement learning algorithms: Q-learning and the Bucket Brigade. We introduce a special case of the Bucket Brigade, and analyze and compare its performance to Qlearning in a number of experiments. The second part of the paper discusses the key problems of reinforcement learning: time and space complexity, input generalization, sensitivity to parameter values, and selection of the reinforcement function. We address the tradeoff be-tween the amount of built in and learned knowledge in the context of the number of training examples required by a learning algorithm. Finally, we suggest directions for future research., |
