In this paper, we present Policy Methods as an alternative to Value Methods to solve Reinforcement Learning problems. The paper proposes a Direct Policy Search algorithm that uses a Neural Network to represent the control policies. Details about the algorithm and the update rules are given. The main application of the proposed algorithm is to implement robot control systems, in which the generalization problem usually arises. In this paper, we point out the suitability of our algorithm in a RL benchmark, that was specially designed to test the generalization capability of RL algorithms. Results check out that policy methods obtain better results than value methods in these situations.
IOS Press, Inc.
6751 Tepper Drive
Clifton, VA 20124
Tel.: +1 703 830 6300
Fax: +1 703 830 2300 email@example.com
(Corporate matters and books only) IOS Press c/o Accucoms US, Inc.
For North America Sales and Customer Service
West Point Commons
Lansdale PA 19446
Tel.: +1 866 855 8967
Fax: +1 215 660 5042 firstname.lastname@example.org