Embodied First Exam: 1,090 embodied tasks test coding agents on 44 robots
wzenus · x · 2026-10-20
Embodied First Exam: putting coding agents to the test in the physical world
Researchers released Embodied First Exam, a benchmark that runs general-purpose coding agents on 1,090 embodied tasks across 31 simulation suites and 44 robots with a single line of code.
Tasks go far beyond pick-and-place:
- Puzzles with spatial constraints
- Olympic sports requiring multi-robot collaboration
- Real-time evaluation, where the world won't pause for the agent
- Dynamic, agile whole-body control
- AutoResearch via sim2sim
The authors observe that agents already handle some physical challenges but still struggle with dexterous manipulation and dynamics-heavy control. Try it: harbor run -d embodied-first-exam -a <harness> -m <model>.
More from Embodied
- mjfit: open-source convex decomposition for tight-clearance parts, 10x faster than SDF — yacineMTB · 2026-10-20
- Waymo reportedly plans to quintuple robotaxi fleet to 20,000 vehicles by end of 2027 — Polymarket · 2026-10-20
- MoCaP: zero-shot mobile manipulation for robots from passive human videos — sherryyangML · 2026-10-20
- Lucky Engine 2026.3 launches: photo-to-simulation for robot policy training — chris_j_paxton · 2026-10-20
- Berkeley EMBER humanoid seminar series: CMU's Guanya Shi on generalist humanoids — GuanyaShi · 2026-10-20
- Same interface drives a real robot: buzz-wire demo and EmbodiedSafetyExam — ManlingLi_ · 2026-10-20