2A2-F24 強化学習法BRLにおけるメタ学習のためのルール発火エントロピの学習収束指標への適用

概要

論文の詳細を見る
A topic of a learning autonomous robot has widely been discussed in these years. Reinforcement learning (RL) is one of the popular methods to this domain. Its performance is, however, quite sensitive to the segmentation of state and action spaces. To overcome this problem, we have been developing a new technique, Bayesian-discrimination-function-based RL (BRL). Furthermore, BRL has proven to be more effective than other standard RL algorithms in dealing with multi-robot system (MRS) problems. However, as most learning systems do, BRL owns the problem of over-fitting. This paper introduces an extended BRL for improving the robustness of MRS. Meta learning based on information entropy of firing rules is adopted for adaptively modifying its learning parameters. Computer simulations are conducted to verify the effectiveness of our proposed method.