Research On Ground Touch Detection Method Of Foot Robot Based On Joint Drive Motor Current

Foot robot has been a hot spot in the field of robot research because of its superior ability to overcome obstacles and flexibility in movement.The walking,jumping,running and other movements of the foot robot are inseparable from the contact with the external environment.In order to realize its stable movement in a complex environment,touchdown detection is very important.This paper presents a research method for touchdown detection of foot robot based on the current of joint drive motor.Firstly,the mathematical model of permanent magnet synchronous motor is established,the modulation principle of Space Vector Pulse Width Modulation is introduced,and the speed and current double closed-loop vector control strategy of i_d=0 is adopted.The relationship between current and driving torque is analyzed,and the collision detection is realized through the change of current.The PMSM vector control system is simulated in Matlab/Simulink.Then,the kinematics model of the one-legged robot is built,and the relationship between the foot trajectory and the joint variables in the reference coordinate system is obtained.Lagrange equation is used to establish its dynamic model,and several joint friction models are proposed.The principle of Gaussian process regression is introduced.Aiming at the influence of factors such as the difficulty of parameter identification,nonlinear friction and error of robot dynamics model,the system probabilistic dynamics model is fitted by using data,and a one-legged robot model based on GPR is established.Secondly,the strategy of learning is based on the state points on the planned motion trajectory rather than the whole workspace,and the angular velocity of multiple cycles is adopted to obtain the training data.The concrete methods of solving covariance function and hyperparameter are discussed.The estimation of touchdown state is analyzed,the modeling strategy of GPR learning is described in detail,and the simulation experiment of model learning is carried out.Finally,the experimental platform of the one-legged robot model is built.The one-legged robot model was used to carry out the aerial phase swing experiment,collision detection experiment,free-falling collision experiment and vertical jump control experiment.The experimental results show that the model based on Gaussian process regrassion can effectively learn the system probabilistic dynamics model of the robot,so as to realize the foot impact state detection based on the joint drive motor current.