Flip Stunts on Bicycle Robots using Iterative Motion Imitation

2026-03-29

AI Summary

This work addresses the challenge of generating kinematically and dynamically feasible motion from initially infeasible reference trajectories by proposing Iterative Motion Imitation (IMI), a method that integrates reinforcement learning with model-based control. Starting from an infeasible reference maneuver, IMI iteratively refines a policy to produce agile and physically realizable front flips. The approach achieves, for the first time without external assistance, successful ground-to-ground and ground-to-table front-flip stunts on the Ultra-Mobility Vehicle bicycle robot platform. Compared to single-stage imitation strategies, the proposed method substantially improves task success rates and demonstrates exceptional robustness and transferability to real-world execution.

Abstract

This work demonstrates a front-flip on bicycle robots via reinforcement learning, particularly by imitating reference motions that are infeasible and imperfect. To address this, we propose Iterative Motion Imitation(IMI), a method that iteratively imitates trajectories generated by prior policy rollouts. Starting from an initial reference that is kinematically or dynamically infeasible, IMI helps train policies that lead to feasible and agile behaviors. We demonstrate our method on Ultra-Mobility Vehicle (UMV), a bicycle robot that is designed to enable agile behaviors. From a self-colliding table-to-ground flip reference generated by a model-based controller, we are able to train policies that enable ground-to-ground and ground-to-table front-flips. We show that compared to a single-shot motion imitation, IMI results in policies with higher success rates and can transfer robustly to the real world. To our knowledge, this is the first unassisted acrobatic flip behavior on such a platform.

Problem

• bicycle robots
• front-flip
• motion imitation
• reinforcement learning
• agile behaviors

Innovation

• Iterative Motion Imitation
• reinforcement learning
• bicycle robot
• acrobatic flip
• motion imitation