NVIDIA unveils VFlash: a System-1 VLM decoding framework for fast visual planning in robots and driving
ZhidingYu · x · 2026-11-06
NVIDIA Research (LPR Team and Jim Fan's GEAR Lab) introduced VFlash, a generalist decoding framework for Physical AI positioned as a System-1 VLM. Core idea: embodied models can think in visual plans — where to grasp, how to move, how a scene evolves — but autoregressive coordinate-token generation makes detailed plans expensive to update continuously. VFlash exploits the known geometric structure of visual primitives (points, boxes, depth, 2D/3D trajectories): a lightweight drafter trained on annotated geometry drafts complete primitives in parallel, achieving faster inference and more accurate robot trajectories than its autoregressive target. Technical report and code are available.
More from Embodied
- GPT-6 Astra tops a 16-benchmark robot index, yet went 0/11 on real plug insertions — AIFlow_ML · 2026-11-06
- Giving Claude a cutting-edge robot body just so it can spoon-feed me — repligate · 2026-11-06
- moonlake previews research on scaling robot learning with simulation in the RL era — du_yilun · 2026-11-06
- Nearly 100 testers praise Sphenomenon MR demo that turns a whole house into voxels — Scobleizer · 2026-11-06
- Spectral Skills lands on real robots: composing motion skills via energy functions — daibond_alpha · 2026-11-06
- Developer gives Claude Desktop a body and sensors via MCP and Raspberry Pi — jfrankie · 2026-11-06