VFlash: Enabling System-1 VLMs for Physical AI · NVIDIA Research
Nvidia Research has introduced VFlash, a generalist decoding framework designed to accelerate visual planning for Physical AI agents such as robots and autonomous vehicles. The method adapts Vision-Language Model speculative decoding by using a lightweight drafter to predict complete visual primitives—including boxes, depth, and trajectories—in a single parallel pass. According to the researchers, VFlash reduces trajectory errors and achieves up to 5x faster robot inference and over 3x faster driving inference on supported hardware. Open article →
- The era of System One Vision Model is here! Introducing VFlash: parsing the visual world in paral...
- VFlash: Fast, Accurate Visual Planning via Flash Primitive Decoding
- NVIDIA’s VFlash speeds robot trajectory decoding 12x, cuts prediction errors by 61%
- VFlash: Parallel decoding for faster, more accurate visual planning VFlash from NVIDIA Research ...
- VFlash: Enabling System-1 VLMs for Physical AI