ReactHuman: a physics-grounded benchmark exposes reactive safety failures in embodied multimodal LLMs
Yizhan Li · hf · 2026-09-15
Yizhan Li released ReactHuman on Hugging Face, a physics-grounded benchmark evaluating multimodal LLMs as decision cores for simulated humanoid robots facing sudden household hazards. The results reveal persistent reactive safety failures: the models struggle to make human-like immediate responses to unexpected physical risks.
More from Embodied
- Pi robot runs fully autonomous for hours at a chocolate factory — stacking boxes was the hard part — chelseabfinn · 2026-09-15
- Peking University Releases MobileVLA-R1 2.0 with RL-Enhanced Robot Control — PekingUniversity · 2026-09-15
- X Humanoid Debuts Pelican-Sim 1.0, a General World Model Simulator for Embodied AI — X-Humanoid · 2026-09-15
- General Robotics pitches AI to automate parts of robot development — VraserX · 2026-09-15
- Full fruit fly brain map, 166,700 neurons, flies a drone in real time — anselm · 2026-09-15
- GenRobot AI Turns Egocentric Human Videos Into Structured Whole-Body Data for Humanoids — CyberRobooo · 2026-09-15