Embodied First Exam: Li Fei-Fei et al. benchmark coding agents on robots, 34% top success
ChongZzZhang · x · 2026-11-17
- Li Fei-Fei, Pieter Abbeel, Dawn Song and collaborators released Embodied First Exam, a "first exam" for coding agents before entering the physical world — controlling robots directly, writing control code, or training policies.
- Scale: 1,156 tasks across 35 suites on 47 robots, with new suites for spatial puzzles, dexterous manipulation, multi-robot interaction, and whole-body control, plus a realtime setting where the world keeps moving while the agent thinks.
- Results are humbling: best success rate is just 34% on the 456 core tasks.
More from Embodied
- Embodied AI field debate: why VLA work is a 'influencer field' and pretraining is the steepest slope — ChongZzZhang · 2026-11-17
- inanimate's WORKS, first "Promptable Products" AI hardware, launches Nov 19 — genmon · 2026-11-17
- Barbell-tracking app built on custom physics engine: LLMs compute forces in 3 steps — wordgrammer · 2026-11-17
- 100,000 hours of egocentric skill data open-sourced on Hugging Face, free for all — lukas_m_ziegler · 2026-11-17
- Skild AI and Nvidia's Steel solver hits 13x PhysX throughput for robot learning — lukas_m_ziegler · 2026-11-17
- Runway open-sources Praxis-1, a 27B MoE robot policy topping RoboLab-120 — lukas_m_ziegler · 2026-11-17