Research On Biped Robot Walking Based On Reinforcement Learning

With the help of artificial intelligence,bipedal robots can replace humans in labor such as material distribution and hazardous operations,freeing up labor.Biped robots have high degrees of freedom,so they can adapt to complex terrain and have huge development space in the future.The research on the structure of bipedal robots at home and abroad mostly focuses on the design of shock absorption structures,lacking consideration for the joint effects of robots;And there are problems such as poor gait adaptability and low learning efficiency when controlling the walking of bipedal robots.In response to these issues,this article studies the key technologies of bipedal robots.A bipedal robot structure that conforms to human joint functions has been designed,and an adaptive walking control algorithm has been improved to improve the efficiency of adaptive stable gait learning for bipedal robots.The main research content is as follows:By analyzing the current research status and future development trends of bipedal robots both domestically and internationally,a bipedal robot with humanoid knee joint function is proposed,and the mechanical and control systems are designed;Focus on designing a humanoid knee joint structure with locking function and a foot structure with shock absorption function;The statics analysis and modal analysis of the overall structure of the robot are carried out using Ansys to verify the reliability of the designed structure.In order to analyze the gait of the designed biped robot,the kinematics and dynamics models of the biped robot were established.The kinematics model of the biped robot is established by D-H method,and the accuracy of the kinematics model is verified by matlab software analysis;The dynamic modeling of biped robot is established based on space operator algebra(SOA)method,and the walking process is simulated using ADAMS software.The motion ability of biped robot is analyzed,and its motion parameters are obtained.To ensure the stable gait of the robot in an adaptive environment,research is conducted on the robot walking control algorithm.To address the slow speed of obtaining adaptive stable gait for bipedal robots in unfamiliar environments,a gating mechanism and attention mechanism are utilized to improve the action network and value network of the proximal strategy optimization(PPO)algorithm;According to the gait of the robot,an appropriate activation function is selected to make the network output nonlinear.By comparing different control algorithms,the superiority of using the improved PPO algorithm to control the walking of bipedal robots in terms of environmental adaptation and gait learning efficiency is verified.Simulate and conduct prototype experiments on the gait of a bipedal robot prototype based on the ROS system.According to the requirements of walking gait for bipedal robots,complete the design of a robot walking controller based on the improved PPO algorithm;Simulation of adaptive walking of bipedal robots under improved PPO algorithm control in Gazebo;Transfer the simulation strategy to the prototype to complete the prototype experiment,verify the adaptability and stability of the algorithm based on improved PPO in the gait control of bipedal robots,and achieve adaptive gait motion of bipedal robots.