Simon Gerstenecker

Benchmark for closed-loop driving generalization in CARLA, revealing widespread failure modes across seven state-of-the-art methods.

Object-based driving planner achieving SOTA in CARLA, with generalization analysis exposing structural failure modes in learned planners.

Scalable reinforcement-learning planning policies trained with simple reward formulations for autonomous driving.

Sim-to-real reinforcement learning, teleoperated imitation learning, and real-world RL on a physical dexterous robotic hand.