复杂环境下仿蛇机器人的路径规划策略  

Path planning strategy of snake-like robot in complex environment

作　　者：李伟庆 王永娟[1] 高云龙 LI Weiqing;WANG Yongjuan;GAO Yunlong(School of Mechanical Engineering,Nanjing University of Science and Technology,Nanjing 210094,China)

机构地区：[1]南京理工大学机械工程学院,南京210094

出　　处：《兵器装备工程学报》2024年第7期28-37,共10页Journal of Ordnance Equipment Engineering

摘　　要：为完成多障碍物复杂环境中的侦察任务,改善仿蛇机器人的路径搜索能力和提升自主决策能力,本文中提出了一种基于先验知识推理库强化学习的路径规划控制策略。首先,建立仿蛇机器人的运动模型和交互环境模型;其次,通过建立模糊逻辑先验知识推理库(FLIS)与Soft Actor-Critic(SAC)动作网络组成的动作分层选择模型,将输出动作空间进行离散化的方法来调整运动控制精度,提供奖惩机制指导机器人与环境模型进行交互,实现机器人运动自主决策的持续生成过程。仿真结果表明:在多障碍物环境中所提出的改进算法得到的运动控制策略收敛速度和鲁棒性明显提升,降低了训练探索次数,提高了机器人对复杂环境的适应性;试验验证了训练所生成的策略模型在真实环境中的可行性。In order to complete the reconnaissance task in the complex environment of multiple obstacles,improve the path search ability of the snake-like robot and enhance the autonomous decision-making ability,this paper proposes a path planning control strategy based on prior knowledge inference library reinforcement learning.Firstly,the motion model and interactive environment model of snake-like robot are established.Secondly,by establishing an action hierarchical selection model composed of Fuzzy logic prior knowledge inference system(FLIS)and Soft Actor-Critic(SAC)action network,the output action space is discretized to adjust the motion control accuracy,and a reward and punishment mechanism is provided to guide the robot to interact with the environment model to realize the continuous generation process of robot motion autonomous decision-making.The simulation results show that the convergence speed and robustness of the motion control strategy obtained by the improved algorithm proposed in the multi-obstacle environment are significantly improved,the number of training explorations is reduced,and the adaptability of the robot to the complex environment is improved.The experiment verifies the feasibility of the strategy model generated by the training in the real environment.

关 键 词：仿蛇机器人 强化学习 先验知识 路径规划 自主避障 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]