RoboColosseum: 8 models fail all 5 Unitree G1 tasks across 5,699 real-robot trials
DJiafei · x · 2026-11-05
First results from RoboColosseum: across 5,699 real-world rollouts at 124 sites, 8 models (five VLAs plus three frontier agents) failed to complete any of the five tasks on the Unitree G1 humanoid. Cross-embodiment ranking consistency was only 0.24 — topping one robot's leaderboard says almost nothing about another. The author's takeaway: vendor demo videos aren't lies, just untested; real verification means asking how many tasks, how many runs, on whose robots — likely done by operators paid hourly.
More from Embodied
- USC's WACL teaches robots whole-arm manipulation while the hand stays busy — ebiyik_ · 2026-11-05
- Robotics startups will burn out before profitability — be the last mover — ___Mufasaa · 2026-11-05
- ExMR open-sources single-video full-body 3D motion capture with robot export — camenduru · 2026-11-05
- Sense launches an always-sensing agent for everything that happens in person — Scobleizer · 2026-11-05
- "Please be patient, it's his first day": new-on-the-job bot flubs go viral — venturetwins · 2026-11-05
- MIT's UQ-DAAAM helps robots know when their memory is wrong, boosting accuracy by 5% — MIT News AI · 2026-11-05