Reflex hits 85.7% catch success from RGB-D and chains catch-carry-handoff demos
DJiafei · x · 2026-10-09
The Reflex project page, code and paper are live. In simulation, Reflex achieves 85.7% catch success from onboard RGB-D versus 87.9% with privileged box observations, with plots showing degradation as throws land farther to the side. The team also connects Reflex to the pretrained SONIC walking controller for a full catch → carry → hand over → return demo. Authors span UW, Stanford, USC, NUS, UNC and AI2.
Related event: Reflex Lets Unitree G1 Catch Tossed Boxes in About a Second(8 posts)→
More from Embodied
- Hugging Face LeRobot launches streaming datasets for 100TB+ robot training — YuXiang_IRVL · 2026-10-09
- SoftBank to acquire RAI Institute, the robotics AI lab founded by Boston Dynamics creator Marc Raibert — lukas_m_ziegler · 2026-10-09
- Dex-One2Many: one human video trains a dexterous hand that transfers zero-shot to real robots — furongh · 2026-10-09
- TIER IV open-sources METEOR: one 54M-parameter net runs 12 driving tasks at 15 FPS on Jetson — rsasaki0109 · 2026-10-09
- Reflex teaches a Unitree G1 humanoid to catch boxes thrown by humans — DJiafei · 2026-10-09
- RoboPrompt lets you steer robots by pointing, circling or sketching trajectories — YouJiacheng · 2026-10-09