Research On Gait Planning And Generation Method Of Small Quadruped Robot

Gait is an important factor that affects the stable motion of the quadruped robot.Among the many gait forms of quadruped robots,diagonal gaits have higher stability and better mobility,so diagonal gaits have become a common gait form in the research of quadruped robots.The study of the state stability becomes the key of diagonal gait control.The motion model of a quadruped robot is analyzed in this thesis.The kinematics equations are established.An index to quantify the stability of diagonal gait is presented.A diagonal gait balance method combining posture and foot drop time difference is proposed.The quadruped robot walks diagonally.At the same time,a gait generation method using deep reinforcement learning is suggested to realize the model-free walking of a quadruped robot.The main work of this thesis is as follows1.First of all,in this thesis the developments of quadruped robots at home and abroad are reviewed.The research status of quadruped robot balancing and generation algorithms are summarized.The chapter arrangement of the thesis are introduced.2.First of all,in this thesis the regular gait of the quadrupeds and designs the topology of the quadruped robot are elaborated.Then,the kinematics model and single leg model of the quadruped robot are established based on the D-H method.According to the single-leg model,the single-leg forward kinematics equation and the single-leg inverse kinematics equation are established.The foot end workspace of the robot is calculated according to the kinematics equations of the robot and the actual rotation angle range of each joint in the model.3.First of all,dynamic analysis based on the spring inverted pendulum(Spring Loaded Inverted Pendulum,SLIP)model is performed to understand the dynamic gait motion characteristics of the quadruped robot.Then the realization method of diagonal gait is implemented according to its characteristics.The foot trajectory equation and curve are solved.Finally,a virtual prototype and simulation environment of a quadruped robot are constructed under Webots.The robot's forward speed,lateral speed,torso height and attitude angle and other data information are recorded in the robot's gait simulation experiment for the next study.This simulation model lays the foundation for gait control algorithm and gait generation algorithm.4.Firstly,in this thesis we analyze the dynamic stability of the quadruped robot.The conditions of the body's dynamic motion stability are summarized Then a quantitative standard is proposed for the stability of the diagonal gait of a quadruped robot.A real-time posture-based feedback strategy is constructed to improve the stability of the body.The effectiveness of the method is verified.Finally,combined with the real-time attitude feedback strategy,a self-adjusting method is proposed based on the center of gravity of the time difference between the corners.Using collecting attitude data in simulation experiment the effectiveness of the control method is verified to improve the stability of the robot through the change of the trunk attitude.5.Firstly,the principle of reinforcement learning is introduced.The framework and process of Deep Q Learning(DQN)algorithm are presented Then combining the DQN algorithm with the quadruped robot controller the quadruped robot DQN controller is trained.Finally,the algorithm's reward function and various parameters are designed to train the quadruped robot in the Webots simulation environment.Using the controller obtained by the training,the robot can achieve smooth motion.The steady motion shows the effectiveness of the proposed method.This provides new ideas for the research on gait generation of quadruped robots.6.The main research work are summarized in this thesis as well as the shortcomings of the proposed method.The direction of future research is discussed.