基于深度强化学习的移动机器人动态路径规划算法
2023,31(1):153-159
摘要：为了在复杂舞台环境下使用移动机器人实现物品搬运或者载人演出，提出了一种基于深度强化学习的动态路径规划算法。首先通过构建全局地图获取移动机器人周围的障碍物信息，将演员和舞台道具分别分类成动态障碍物和静态障碍物。然后建立局部地图，通过LSTM网络编码动态障碍物信息，使用社会注意力机制计算每个动态障碍物的重要性来实现更好的避障效果。通过构建新的奖励函数来实现对动静态障碍物的不同躲避情况。最后通过模仿学习和优先级经验回放技术来提高网络的收敛速度，从而实现在舞台复杂环境下的移动机器人的动态路径规划。实验结果表明，该网络的收敛速度明显提高，在不同障碍物环境下都能够表现出好的动态避障效果。
关键词：移动机器人；LSTM；深度强化学习；动态路径规划；实时避障
Dynamic path planning algorithm of mobile robot based on deep reinforcement learning
Abstract：A dynamic path planning algorithm based on deep reinforcement learning is proposed in order to use mobile robots to carry goods or perform manned performances in complex stage environment. Firstly, the obstacle information around the mobile robot is obtained by constructing a global map, and the actors and stage props are classified into dynamic obstacles and static obstacles respectively. Then establish a local map, encode the dynamic obstacle information through LSTM network, and calculate the importance of each dynamic obstacle through social attention mechanism to achieve better obstacle avoidance effect. By constructing a new reward function, different avoidance situations of dynamic and static obstacles are realized. Finally, simulation learning and priority experience playback technology are used to improve the convergence speed of the network, so as to realize the dynamic path planning of mobile robot in the complex stage environment. The experimental results show that the convergence speed of the network is significantly improved, and it can show good dynamic obstacle avoidance effect in different obstacle environments.
Key words：mobile robot; LSTM； deep reinforcement learning; dynamic path planning; real time obstacle avoidance
收稿日期：2022-06-11
基金项目：国家自然科学基金项目 （61973275）
