ForeAct：以高效视觉预见规划引导视觉语言动作模型
ForeAct: Steering Your VLA with Efficient Visual Foresight Planning
把「想象未来观测」作为可插拔的视觉输入增强，让 VLA 从高层语义推理中解放出来专注视觉-运动推理；其 0.33s 生成 640x480 预见图像的轻量生成模块与百万级跨具身预训练思路，可直接迁移到任何以图像为输入的策略模型上。
原文依据 state-of-the-art VLAs can integrate our planner seamlessly by simply augmenting their visual inputs, without any architectural modification
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论视觉语言动作（VLA）模型将高层语言指令转化为具体可执行的动作，这一任务在开放世界环境中尤为困难。本文提出视觉预见规划（ForeAct），一种通用且高效的规划器，利用想象出的未来观测与子任务描述逐步引导 VLA。借助想象出的未来观测，VLA 可以专注于视觉-运动推理而非高层语义推理，从而提升准确性与泛化能力。该规划器包含一个高效的预见图像生成模块，可在 H100 GPU 上仅用 0.33 秒从当前视觉输入与语言指令预测出高质量的 640x480 未来观测；同时包含一个视觉语言模型，对任务进行推理并为生成器与 VLA 生成子任务描述。重要的是，最优的 VLA 只需增强其视觉输入即可无缝集成该规划器，无需任何架构修改。预见生成器在超过 100 万条多任务、跨具身片段上预训练，从而学习到鲁棒的具身动力学。我们在一个由 11 个多样化、多步骤真实世界任务构成的基准上评估该框架，取得 87.4% 的平均成功率，相较 π_0 基线（46.5%）取得 +40.9% 的绝对提升，相较引入文本子任务引导的 π_0（57.1%）取得 +30.3% 的绝对提升。
英文原文摘要 1487 字符
Vision-Language-Action (VLA) models convert high-level language instructions into concrete, executable actions, a task that is especially challenging in open-world environments. We present Visual Foresight Planning (ForeAct), a general and efficient planner that guides a VLA step-by-step using imagined future observations and subtask descriptions. With an imagined future observation, the VLA can focus on visuo-motor inference rather than high-level semantic reasoning, leading to improved accuracy and generalization. Our planner comprises a highly efficient foresight image generation module that predicts a high-quality 640x480 future observation from the current visual input and language instruction within only 0.33s on an H100 GPU, together with a vision-language model that reasons over the task and produces subtask descriptions for both the generator and the VLA. Importantly, state-of-the-art VLAs can integrate our planner seamlessly by simply augmenting their visual inputs, without any architectural modification. The foresight generator is pretrained on over 1 million multi-task, cross-embodiment episodes, enabling it to learn robust embodied dynamics. We evaluate our framework on a benchmark that consists of 11 diverse, multi-step real-world tasks. It achieves an average success rate of 87.4%, demonstrating a +40.9% absolute improvement over the \pi_0 baseline (46.5%) and a +30.3% absolute improvement over \pi_0 augmented with textual subtask guidance (57.1%).