RoboDojo Benchmarks 30+ Embodied AI Models: General Robotics Still Unstable

A new benchmark called RoboDojo evaluated over 30 frontier embodied AI models, concluding that general robotic policies are still far from robust in real-world operations. Tencent's open-source Hy-embodied VLA currently tops the leaderboard, while pi-0.5 also shows strong performance.

2026-07-09 ~ 2026-07-10 · 2 related posts

• Evaluating 30+ Embodied AI Models Reveals Lack of Robustness — ZeYanjie · 2026-07-09
• Embodied AI Leaderboard and RoboDojo Eval — furongh · 2026-07-10