论文HF Daily Papers · arXiv 2611.08449昨天

VFlash视觉规划解码提速25倍，机器人任务成功率升至70.2%

针对具身智能视觉规划生成慢痛点，VFlash通过Flash Primitive Decoding联合预测坐标，3D点运动 forecasting 提速约25倍。在MolmoSpaces四项任务中平均成功率达70.2%，较自回归基线提升7.6个百分点。

AI 整理

感知与导航
具身模型

本文由 AI 根据公开内容整理，可能有误，请以原文为准：VFlash: Fast, Accurate Visual Planning via Flash Primitive Decoding（HF Daily Papers · arXiv 2611.08449）

核心内容

针对物理AI任务中自回归视觉语言模型逐token生成空间规划效率低的问题，研究提出VFlash方法。该方法利用视觉原语结构特性，通过轻量级drafter单次前向传播联合预测完整原语坐标，支持变长轨迹与紧凑框。实测显示其在3D点运动预测任务中生成速度较自回归解码提升约25倍，且在2D/3D机器人轨迹任务中预测误差持续降低。

VFlash训练直接基于标注原语，无需复现目标token序列即可接受结构有效预测，解耦了空间精度与文本对齐约束。在世界动作模型验证中，以VFlash规划为条件的系统在MolmoSpaces四个任务套件上取得70.2%平均成功率，显著优于基线模型的62.6%。该架构适配不同原语类型与维度，为高频实时机器人控制提供了高效视觉规划后端方案。

关键看点

解码加速：采用Flash Primitive Decoding联合预测坐标，3D点运动forecasting生成速度较自回归解码提升约25倍
任务性能：条件世界动作模型在MolmoSpaces四套件平均成功率70.2%，较自回归基线62.6%提升7.6个百分点
训练解耦：直接基于标注原语训练，接受结构有效预测而无需目标token一致，降低空间推理对文本对齐依赖
原语适配：轻量级drafter单次前向传播生成完整原语，动态适配类型、维度及变长轨迹，兼容2D/3D规划

工程与产业观察

VFlash将视觉规划从串行文本生成转为结构化并行解码，25倍提速与7.6%成功率增益验证了原语级联合预测在具身控制中的工程价值。其解耦token对齐的训练范式，为构建低延迟、高精度的端到端机器人策略模型提供了可复用的加速解码参考。

英文摘要 (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. 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.

Hugging Face Daily Papers 热度：60 upvotes 作者：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 Yu…