ForeAct: Steering Your VLA with Efficient Visual Foresight Planning

TL;DR

ForeAct introduces an efficient visual foresight planning approach that enhances vision-language-action models by predicting future observations to improve accuracy and generalization in open-world tasks.

Contribution

The paper presents a novel foresight planning module that predicts future observations, enabling better visuo-motor inference without architectural changes to existing VLAs.

Findings

Achieves 87.4% success rate on real-world tasks

Improves performance by over 40% compared to baseline

Foresight generator trained on 1 million episodes

Abstract

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 640$\times$480 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%).