基于改进强化学习算法的移动机器人路径规划研究

PATH PLANNING FOR MOBILE ROBOT BASED ON IMPROVED REINFORCEMENT LEARNING ALGORITHM

针对强化学习中的标准Q-learning算法应用在路径规划中存在的计算效率低的问题,提出一种改进Q-learning算法。改进后的Q-learning算法在原来标准Q-learning算法的基础上增加了一层深度学习层并且在算法初始化的过程中加入了关于环境的先验知识作为启发信息,从而避免了学习前期探索的盲目性,有效地提高了算法计算效率。通过与标准Q-learning算法、增加深度学习层的Q-learning算法、引入人工引力场的Q-learning算法、深度双Q网络相比较,改进后的Q-learning算法在小维度的环境下具有更高的计算效率。

Aiming at the problem that the standard Q-learning algorithm in reinforcement learning has low calculation efficiency in path planning, we propose an improved Q-learning algorithm. The improved Q-learning algorithm added a deep learning layer on the basis of the original standard Q-learning algorithm. It adopted the prior knowledge about the environment as the heuristic information in the process of algorithm initialization, so as to avoid the blindness of exploration in the early stage of learning and effectively improve the calculation efficiency of the algorithm. Compared with the standard Q-learning algorithm, Q-learning algorithm with deep learning layer, Q-learning algorithm with artificial gravitational field and deep double Q network, the improved Q-learning algorithm has higher computational efficiency in small-dimensional environment.