VFlash：通过 Flash Primitive Decoding 实现快速准确的视觉规划
原标题：Paper page - VFlash: Fast, Accurate Visual Planning via Flash Primitive Decoding
AI 摘要
研究者提出 VFlash，一种通过 Flash Primitive Decoding 实现快速、准确视觉规划的方法。该方法利用视觉基元（点、边界框、轨迹）的已知结构，用轻量级 drafter 在一次前向传播中联合预测整个基元的坐标，而非逐 token 自回归生成。在 2D 和 3D 机器人轨迹任务上，VFlash 持续降低预测误差，3D 点运动预测的初步单次 rollout 测量显示生成速度约提升 25 倍；以 VFlash 规划为条件的世界动作模型在四个 MolmoSpaces 任务套件上平均成功率达 70.2%，基线为 62.6%。
正文节选
VFlash: Fast, Accurate Visual Planning via Flash Primitive Decoding 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 v