Probabilistic Dynamic Modeling and Control for Skid-Steered Mobile Robots in Off-Road Environments

Abstract

Skid-Steered Mobile Robots (SSMRs) are commonly deployed for autonomous navigation across challenging off-road terrains due to their high maneuverability. However, modeling the tire-terrain interactions for these robots when operating at their dynamic limits is challenging, since slipping and skidding govern their movement. During nominal operation, the data collected from the deviation of the robot’s measured states from their commanded values can be informative of these hard-to-model dynamics. In this work-in-progress paper, we propose a probabilistic motion model for SSMRs by leveraging least squares and Sparse Gaussian Process Regression (SGPR) algorithms. This model allows for a nonlinear stochastic Model Predictive Control (MPC) formulation that can be solved in real-time. Initial results on the application of GPR to account for unmodeled dynamics of a physics-simulated quadrotor are shown, suggesting that it can similarly be put to good use for off-road autonomy applications. We explain how these results reinforce the promising application of an SGPR model to risk-averse motion planning for SSMRs.