The Strategy Entropy of Reinforcement Learning for Mobile Robot Navigation in Complex Environments

In this paper, the concept of entropy is introduced into reinforcement learning for mobile robot control. The definitions of the local and global strategy entropy are proposed respectively. The global strategy entropy is proved to be a quantitative problem-independent measurement for the learning progress, i.e. the convergence degree of the strategy. To improve the learning performance, a new learning algorithm with self-adaptive learning rate is proposed based on the local strategy entropy. Robot navigation in multi-obstacle environments is achieved with the proposed learning algorithm. The experimental results show that learning based on the local strategy entropy has better learning performance than learning with fixed learning rates.