RoboColosseum benchmarks robot policies on real hardware across 20,000+ blind runs
DJiafei · x · 2026-11-06
Skipping simulated leaderboards, the new RoboColosseum benchmark evaluates robot policies via blind tests on real hardware like the Unitree G1, spanning 150+ varied environments and 20,000+ real-world runs — with every rollout logged openly on Hugging Face, including failure trajectories.
Key finding: policy superiority flipped dramatically based on robot morphology and task nuance rather than model parameter count. Leaderboard snapshots show top VLA policies (NVIDIA GROOT-M17, Physical Intelligence π05, Galaxea G05, Ai2 MolmoAct2) at 0% full success on G1 tasks, while π05-YAM hits 94% on the YAM platform. Academic teams can apply for 10,000 sponsored evaluation hours.
More from Embodied
- Hugging Face co-launches RoboColosseum: open evals on 100+ real robots — DJiafei · 2026-11-06
- RoboColosseum details: 5 VLAs ranked across 20,000+ real-world runs — DJiafei · 2026-11-06
- University of Michigan's 26-lecture robotics math course (ROB 501) is free on YouTube — techNmak · 2026-11-06
- Musk: Tesla to wind down Cybertruck production mid-next year to boost Cybercab output — elonmusk · 2026-11-06
- Xinhua hires two UBTECH UWORLD U1 humanoid robots as news anchors — CyberRobooo · 2026-11-06
- SLA scales spatial-language-action models via continual pre-training with frozen 3D alignment — HKUSTGZ · 2026-11-06