基于Soft-Actor-Critic算法的机器人局部路径规划算法

Robot Local Path Planning Algorithm Based on Soft-Actor-Critic Algorithm

近年来移动机器人应用逐渐广泛,以定位、路径规划等导航技术成为移动机器人研究的热点问题,随着移动机器人执行任务的环境复杂度逐渐增加,移动机器人通过与环境交互实时学习进行路径规划,成为新的研究趋势。作者提出了一种局部路径规划算法,以Soft-Actor-Critic(SAC)算法为框架,以实现机器人通过激光雷达获取的地图信息进行局部路径规划。首先,针对规划问题设计连续的状态-动作变量,并设计了一种连续的奖励函数,使得移动机器人每采取一个动作都可以获得相应的奖励,提高了训练效率,最后建立仿真环境,对智能体进行训练学习,结果验证了算法的有效性。

Mobile robot applications in recent years has been to localization,path planning for mobile robot navigation technology such as the hot issues of the study,along with the mobile robot to perform a task environment complexity increases gradually,mobile robot path planning,through the interaction with the environment in real time to study to become a new research trend,in this paper,we propose a framework for training Soft-Actor-Critic(SAC)algorithm,in order to realize the robot with a laser radar map information for local path planning,first for planning problems involving statevariable,and designed a kind of compensation function in a row,Finally,a simulation environment is established to train and learn the agent,and the results verify the effectiveness of the algorithm.