RADAR-Based Safe Pull-Over of Autonomous Racing Cars in Localization Failure Scenarios

This paper presents a RADAR-Based, vehicle lateral dynamics control algorithm that ensures a safe pull-over of autonomous racing vehicles to the roadside barriers in the event of localization failures. The position and curvature of the roadside barriers are estimated from RADAR measurements through a Total Least Squares algorithm and fed to a Linear-Quadratic-Regulator (LQR), which outputs the steering commands to the vehicle. As the estimates accuracy varies with the distance to the barriers, the proposed algorithm controls the car at a target distance which is dynamically adjusted based on the confidence interval of the barrier distance estimate. The convergence of the resulting closed-loop, nonlinear system to a minimum safe distance to the barrier is proven through Lyapunov stability theory. We consider the setup based on the Dallara AV-21, a fully autonomous racing car competing in the Indy Autonomous Challenge. Simulations on the Las Vegas Motor Speedway track demonstrate the emergency controller's effectiveness, offering a robust safety solution for racing cars in case of localization losses.