LineRides: Line-Guided RL for Bicycle Robot Stunts

Researchers have developed LineRides, a line-guided reinforcement learning framework that enables a custom bicycle robot to perform stunts without demonstrations or explicit timing. The framework uses a user-provided spatial guideline and sparse key-orientations to train the robot, handling physically infeasible guidelines with a tracking margin. Tested on the Ultra Mobility Vehicle (UMV), the policy supports seamless transitions between normal driving and stunt execution. The work addresses the challenge of designing reward functions for agile robotic maneuvers, particularly for novel platforms where reference motions are unavailable.