Torc Robotics published a draft paper titled "Beyond Imitation Learning: How E2E Self-Play Reinforcement Learning Scales for Autonomous Driving," describing a method that uses self-play DAgger to distill a privileged reinforcement learning driving policy into a camera-based policy1. The paper is listed for presentation at CoRL 2026 in Austin, TX, November 9–12.
Torc Robotics Details Self-Play RL Method for Autonomous Trucking
Torc Robotics draft paper describes a self-play DAgger method distilling a privileged RL driving policy into a camera-based policy, set for CoRL 2026.