Show-Harness lets a VLM drive real robots via a physical interface
jiqizhixin · x · 2026-10-05
NUS Showlab presents Show-Harness: instead of retraining a bigger robot policy, give a vision-language model a Computer-Use-style physical interface so it can directly operate real robots.
- Background: Frontier foundation models hold the strongest visual understanding, spatial reasoning, and task decomposition, but these capabilities don't naturally translate into robot actions.
- Problems with prior approaches: Traditional VLAs regress continuous control from images and instructions, requiring re-adaptation per robot and data distribution; systems where the LLM only plans and hands off to a separate controller widen the gap between semantic decisions and physical execution.
- Approach: Show-Harness takes a third path — a physical interface the VLM can understand, compose, and continuously correct, turning Robot-Use into something like Computer-Use.
More from Embodied
- Zoox robotaxi halts in Vegas traffic despite 3 lidars saying the road was clear — jamesdouma · 2026-10-05
- LinkerHand O30 dexterous hand performs card tricks, team to reveal how it was done — DJiafei · 2026-10-05
- Webcam-to-Pen-Plotter Demo at AI Tinkerers Tokyo: Photo to Ink Drawing via AxiDraw — Stefania_druga · 2026-10-05
- Argentine hacker builds robot arm to automatically baste his asado in San Francisco — StewartalsopIII · 2026-10-05
- Paralyzed Designer Creates Figurine via Brain-Computer Interface, AI, and 3D Printing — yogthos · 2026-10-05
- Waymo's new rides keep getting stuck: user spots 3 robotaxis blocking a narrow street — var_epsilon · 2026-10-05