Research On Walking Stability Of Biped Robot Based On Improved DDPG Algorithm

Biped robot is a research hotspot and an important research direction in the field of legged robots.Compared with other types of wheeled and tracked robots,it has higher flexibility and environmental adaptability,and it's closer to human characteristics with important research value.Walking stability is the foundation and key for the biped robot to be used in the actual environment in the future.The dissertation takes the physical prototype of a 12-degree-of-freedom biped robot as the research object,and studies the gait planning problem of the biped robot.The main work content includes the following aspects:(1)Structural design and kinematics analysis of biped robot.Study the structure of human legs and walking,and design the structural characteristics of biped robots based on the structural characteristics of human legs.Based on the DH coordinate system,the positive kinematics equation of the biped robot is constructed,and the numerical solution method of inverse kinematics is given,which lays a theoretical foundation for the subsequent gait planning.(2)Biped walking process analysis and offline gait planning.Introduce the ZMP stability criterion for testing the dynamic balance of the biped robot,analyze the biped walking process and laws,design the related walking parameters,and use interpolation algorithm to plan the biped in the forward and lateral planes at the starting,stopping,and periodic walking stages respectively The trajectory curve of each joint of the robot.(3)Self-learning feedback control of biped robot.The deep deterministic policy gradient algorithm(DDPG algorithm)in reinforcement learning is applied to the optimization of the walking posture of the biped robot to cope with the uncertainties caused by various interference factors during the walking process,and the RBF neural network is targeted network and Sum Tree sample pool tools to improve the efficiency of the improved DDPG algorithm and enhance the environmental adaptability of biped robot walking.(4)Biped robot gait simulation and biped stable walking experiment.According to the built bipedal robot prototype model,the corresponding three-dimensional model is created in Solidworks,and the SWURDF plug-in is used to transfer to the ROS platform,and the dynamic simulation is performed in the Gazebo simulation engine.Then the control system and drive system needed for the experiment are designed,and the corresponding hardware is configured to complete the experiment,and the rationality and correctness of the walking method are verified.Theoretical research,simulation analysis and experimental results show that the offline planning-online adjustment biped robot walking control method proposed in this topic has the advantages of high efficiency,good stability and strong adaptability.The research of this subject also provides reference and basis for further application of intelligent technologies such as reinforcement learning to the traditional control field in the future.