Reproduce “MOBIUS: A Multi-Modal Bipedal Robot that can Walk, Crawl, Climb, and Roll” 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
Alexander Schperberg, Yusuke Tanaka, Stefano Di Cairano, Dennis Hong
MOBIUS is a multi-modal bipedal robot developed by researchers from MERL, ETH Zurich, and UCLA, capable of walking, crawling, rolling, and dynamic free climbing using a single, non-reconfigurable morphology. It successfully performs pinch-grasp pull-ups supporting its full body weight, leverages reinforcement learning for robust locomotion, and employs a multi-modal planner that shows crawling is more energy-efficient (65 J/m) than bipedal movement (104 J/m).
Key insights
- Tightly integrating hardware morphology with advanced control and planning architectures enables a single robotic platform to achieve a wide range of diverse and challenging multi-modal tasks, including dynamic free climbing.
- Extensive domain randomization in simulation is crucial for successful zero-shot sim-to-real transfer of reinforcement learning policies, leading to significantly higher robustness against external disturbances in bipedal locomotion.
- A high-level planner combining path planning with discrete mode selection can optimize for exploration, goal reaching, and energy efficiency, revealing the inherent trade-offs between different locomotion modes (e.g., crawling's energy efficiency versus rolling's speed).
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.