Research On Analysis And Controller Design Of Periodic Gaits For Biped Robot

As robot technology continuously develops,the application of robots has increasingly penetrated more corners of industrial fields and everyday life.Biped robots have attracted extensive attention of academic and industrial community because of its high flexibility and strong adaptability to the environment.Gait analysis and control are two important topics in the field of biped robot research,but there still exists some open problems in these topics,such as energy efficiency and stability of gaits.This thesis aims to study the biped robot in two-dimensional and three-dimensional,specifically to analyze its periodic dynamic gait and design nonlinear controllers,with the intention to obtain better and more stable biped gaits.The specific research contents of this thesis are listed below.Firstly,the two-dimensional and three-dimensional dynamics models of biped robot are established.For the biped robot in the two-dimensional space,the foot contact of biped robot with the environment is modeled as an elastic contact,thus the discontinuous state of the robot caused by environmental impact is eliminated,and a continuous two-dimensional spatial dynamic model of biped robot is constructed,and then its dynamic equation is deduced.For the biped robot in the three-dimensional space,a relatively complex three-dimensional biped model is established,which respects a more complex robot physical structure and its simulation environment is constructed.The two models will be applied to the gait analysis and controller design of biped robot in two-dimensional and three-dimensional space.Secondly,based on the two-dimensional spatial dynamics model of biped robot,a class of energy-saving periodic gait patterns,called natural oscillation gait,is analyzed.Based on the MHB(multivariable harmonic balance)method,the natural oscillation gait of closedloop feedback system is effectively analyzed and predicted.Then,a closed-loop controller for natural oscillation gait based on CPG(Central Pattern Generator)is proposed.Based on MHB,the stability criterion of gait controller can be explicitly established.The simulation results show that the designed control method achieves a stable rhythmic natural oscillation gait for the biped robot.Finally,based on the three-dimensional model of biped robot,in order to mitigate the disadvantages of reinforcement learning algorithm,such as low sampling efficiency and difficulty in searching policies for desired gaits,a new reward function is designed to improve reinforcement learning algorithm.Due to the periodic characteristics of biped gait,a cyclic index time variable is introduced to describe the periodicity,and then a periodic reward function is designed.Based on the new reward function,a new reinforcement learning control method is designed to derive the optimal policy for biped gaits.Simulations are conducted to verify the effectiveness of the new reinforcement learning control method for realizing the preset gait pattern.