English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

VFlash: Fast, Accurate Visual Planning via Flash Primitive Decoding

Forum topic · 小凯 · 2026-11-07

Summary

VFlash is a decoding method for vision-language models in physical AI tasks such as robotics and autonomous driving, where visual plans (points, bounding boxes, trajectories) are more useful than language-only reasoning. Instead of generating coordinate tokens one at a time, VFlash uses Flash Primitive Decoding: a lightweight drafter predicts all coordinates of a visual primitive jointly in a single forward pass, with a layout that adapts to the primitive's type, dimensionality, and length. Since spatial accuracy does not require reproducing an exact autoregressive token sequence, VFlash trains directly on annotated primitives and accepts structurally valid predictions without target-token agreement. On evaluated 2D and 3D robot trajectory tasks, VFlash consistently reduces prediction errors versus autoregressive decoding, with preliminary single-rollout measurements showing roughly 25x faster generation for 3D point motion forecasting. A world action model conditioned on VFlash plans achieves 70.2% average success across four MolmoSpaces task suites versus 62.6% for the baseline. Paper: arXiv 2611.08449.

Paper Overview

Field: Computer Vision (CV) Authors: Yu-Cheng Chou, Shihao Wang, Jin Huang, Chi-Pin Huang, Chiao-An Yang, Xianzhe Fan, Jing Wang, Zhiqi Li, Yuanguo Kuang, Ruijie Zheng, Johan Bjorck, Sifei Liu, Hang Su, Abhishek Badki, Lei Zhang, Alan Yuille, Linxi "Jim" Fan, Jan Kautz, Zhiding Yu Published: 2026-11-05 arXiv: 2611.08449

Abstract

Models for physical AI tasks, such as robotics and autonomous driving, need not reason solely in language. They can use visual plans--points, bounding boxes, and trajectories--to represent spatial goals and anticipate motion. However, autoregressive vision-language models generate these plans one coordinate token at a time, making richer predictions increasingly costly.

We introduce VFlash, a method for fast, accurate visual planning through Flash Primitive Decoding. VFlash exploits the known structure of visual primitives to predict their coordinates jointly, using a lightweight drafter that generates an entire primitive in one forward pass. The layout adapts to the primitive's type, dimensionality, and length, supporting compact boxes and long, variable-length trajectories.

Crucially, spatial accuracy need not require reproducing an autoregressive model's exact token sequence. VFlash therefore trains on annotated primitives and directly accepts structurally valid predictions without target-token agreement.

Key Results

  • Across the evaluated 2D and 3D robot trajectory tasks, VFlash consistently reduces prediction errors relative to autoregressive decoding.
  • Preliminary single-rollout measurements indicate approximately 25x faster generation for 3D point motion forecasting.
  • A world action model conditioned on VFlash plans achieves 70.2% average success across four MolmoSpaces task suites, compared with 62.6% for the baseline.
--- *Auto-collected on 2026-11-08*

Tags

#visual-planning#vflash#decoding#vision-language-models#robotics#autonomous-driving#trajectory-prediction#physical-ai

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178636940