Benchmarking Empirical and Learning-Based Approaches for Feedforward Steering Control in Autonomous Racing

Researchers benchmarked two learning-based and two analytical feedforward steering controllers for autonomous racing in a high-fidelity simulator.

The learning-based methods achieved the best open-loop prediction accuracy, but that did not translate into the best driving performance.

The empirical EHD method performed best in closed-loop testing, delivering the strongest robustness and fastest lap times.

The result suggests autonomous racing controllers should be judged by full-stack racing performance, not prediction error alone.