Robust Autonomous Driving Training

closed-loop min-max training and learnability-guided curricula for long-tail robustness.

Beyond finding failures, I am interested in turning adversarial scenarios into useful training signals for autonomous driving policies.

Recent work formulates closed-loop adversarial training as a min-max game and studies how learnability can guide curricula for safer and more efficient policy improvement.

Representative work: ADV-0 and AlignADV.