Trajectory Prediction Based On Estimation Of Instantaneous Centers Of Rotation In Real Time For Skid-steer Vehicles

Skid-steer vehicles change their motion direction by adjusting both sides rolling speeds,and they are widely used in different areas such as agriculture,military,forestry,mining,and planetary exploration due to their simple mechanism,high maneuverability and strong traction.An abstract vehicle model is needed to predict the vehicle's motion in the local planner of automatic driving vehicles.However,the skid steering characteristics makes it difficult to obtain an accurate motion prediction.Therefore it is important to study the estimation of the slip parameters and its influence on the vehicle movement,which makes significant sense for the realization of the unmanned vehicle.In this paper,we study motion prediction based on estimation of instantaneous centers of rotation in real time for skid-steer vehicles.A steering kinematics modeling method based on the instantaneous center of rotation is presented in the thorough analysis to the sliding steering characteristics of kinematics and dynamics.By linearization of the differential equation of the vehicle motion,a linear time varying perturbation matrix differential equation is established,and the closed analytical solution of the trajectory error is derived based on the linear control theory.Based on the closed analytical solution,the non-system pose error of the vehicle based on the covariance matrix is further analyzed.Then the nonlinear least square method based on Levenberg-Marquardt algorithm and the extended Kalman filter method are used to estimate the slip parameters in real time and the future motion trajectory is predicted for a few seconds given a set of input commands.The performance of the algorithm is analyzed by simulation experiments.The simulation results show that although the estimated slip parameters based on L-M algorithm are not necessarily convergence to the global minimum,the predicted pose errors using this method greatly decrease than using the traditional method.The estimated slip parameters based on EKF can quickly converge to the true values,which enable the prediction pose errors are greatly reduced.Meanwhile,the simulation results show that when using low sampling frequency sensor,the proposed method of integrating the differential equations of motion is superior to the traditional method.Finally,the real vehicle tests based on a tracked vehicle platform and a 6×6 wheeled vehicle platform are carried out and the two algorithms are verified.The results show that,compared with the traditional method,the predicted pose errors based on the L-M method can be significantly reduced.For the tracked vehicle platform,the errors are reduced by more than 40% and for the 6×6 wheeled vehicle platform,the reduction of pose errors are about 60%.Compare with ignoring the slippage,the predicted pose errors using EKF method are also significantly reduced.For the tracked vehicle platform,the predicted position and heading errors are reduced by about 76.5% and 74.4%;for the 6×6 wheeled vehicle platform,the predicted position and heading errors are reduced by about 86.6% and 90.6%.