Reproduce “Learning to Walk and Fly with Adversarial Motion Priors” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Giuseppe L'Erario, Drew Hanover, Angel Romero, Yunlong Song, Gabriele Nava, Paolo Maria Viceconte, Daniele Pucci, Davide Scaramuzza

A learning-based method leveraging Reinforcement Learning and Adversarial Motion Priors enables an aerial humanoid robot to autonomously execute smooth, energy-efficient transitions between legged and aerial locomotion. The robot adaptively switches modes based on environmental context and terrain, achieving both human-like walking and optimized flying behaviors.

Key insights

  • Adversarial Motion Priors enable the spontaneous emergence of multimodal locomotion and adaptive mode switching without explicit high-level planners or state machines.
  • The integration of both walking (human-like) and flying (optimized) motion priors is crucial for achieving efficient, versatile, and natural-looking multimodal behaviors.
  • A thrust penalization term in the reward function effectively guides the robot towards energy-efficient locomotion, preferring walking over flying when feasible.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.