一种基于性能势的无折扣强化学习算法

Undiscounted Reinforcement Learning Algorithm Based on Performance Potentials

传统基于性能势的学习算法能获得马尔可夫决策问题的最优策略。这些算法主要采用单路径采样的方法,使得学习算法效率不高。将性能势与强化学习相结合,提出了一种基于性能势的无折扣值迭代学习算法——G学习,并将其与经典的无折扣强化学习算法(R学习)相比较,获得了较好的实验结果。

Traditional performance potential-based learning algorithms can obtain optimal policies in MDP problems. They mainly adopt single sample path based on methods which make them less efficient. In this paper,a new learning algorithm which utilizes performance potential and reinforcement learning is proposed. Compared with the classic R-learning algorithm ,it has promising results.