Hugging Face to release all RoboColosseum robot eval rollouts, including failures

micoolcho · x · 2026-11-13

Wrapping up CoRL, Hugging Face's Pepijn announced that every RoboColosseum eval rollout will be released, including failures. The benchmark's organizers stress that while an LLM can be benchmarked from a laptop, robot policies require real hardware and environments — RoboColosseum provides real robots operated by independent operators in new scenes, since domain experts need to see how models actually behave.

Original post →

More from Embodied

Embodied channel →