Learning Based Trajectory Planning and Safe Obstacle Avoidance for Tracked Vehicle Under Imprecise Model

A trajectory planning method based on Gaussian regression for Model Predictive Control is proposed to address the safety obstacle avoidance problem of tracked vehicle on unstructured roads. The proposed method takes into account the parameter uncertainty caused by different road conditions and slopes, with the goal of ensuring safe obstacle avoidance. In order to achieve safe obstacle avoidance in imprecise model situations, Gaussian regression process is used to learn the average model error caused by time-varying parameters, and variance is introduced in the design of control barrier function. The characteristic of this method is to increase constraints on the control barrier area to ensure safety. Simulation shows that the designed algorithm can achieve safe obstacle avoidance while reaching the target point.