Safe High-Performance Autonomous Off-Road Driving Using Covariance Steering Stochastic Model Predictive Control

Autonomous racing is a high-performance, safety-critical task that inherently involves a high degree of uncertainty (especially in off-road unstructured environments), as driving conditions can vary and tire-terrain interactions are difficult to model accurately. On the one hand, the vehicle needs to drive fast while, on the other hand, it must avoid crashing, thus requiring a tradeoff between performance and safety. This work develops a stochastic model predictive controller (SMPC) for uncertain systems with additive Gaussian noise subject to state and control constraints and applies it to off-road autonomous racing. The proposed approach is based on the recently developed finite-horizon optimal covariance steering (CS) control theory, which steers the system state's mean and covariance to prescribed target values at a given terminal time. We show that the proposed CS-SMPC algorithm can deal with unbounded Gaussian additive noise while ensuring stability. The effectiveness of the proposed approach is demonstrated via both numerical and experimental tests using a scaled autonomous racing platform, as well as on an actual full-size vehicle during a global positioning system (GPS)-denied autonomous driving task.