# World Action Agent：用 VLM 通过世界动作排练操控机器人

- 来源：HuggingFace Daily Papers（社区热门论文）
- 发布时间：2026-09-24 08:00
- AIHOT 分数：43
- AIHOT 链接：https://aihot.news/items/cmuge8kyh0palrogvhwjdjrq2
- 原文链接：https://arxiv.org/abs/2609.29964

## AI 摘要

World Action Agent（WAA）是一个多智能体框架，让 VLM 借助基础工具在视觉动作工作区中直接操控机器人，工作区具备接触视图、动作排练与视图内校正三项特性。

## 正文

General-purpose vision-language models (VLMs) bring broad knowledge and spatial reasoning to robot manipulation, yet existing systems either use them indirectly, to predict constraints or write programs, or give them a view of the scene rather than a world in which to act. We present World Action Agent (WAA), a multi-agent harness through which VLMs pilot robots with basic tools, making every decision within a visual action workspace. The workspace has three properties. Contact views, selected automatically from the scene geometry, present the scene around the current interaction.

Action rehearsal turns each action into an editable proposal that the agent, alone or through an Imagination Agent, previews and revises against planning feedback before execution. In-view correction closes the loop between observation, rehearsal, and low-level execution, letting the agent remove residual offsets in the view where it observes them. Through the same workspace, WAA acquires embodied procedural knowledge in two ways: it evolves multimodal skills from expert videos and human teaching under evidence-based review and consults them through a Skill Agent, and its interaction traces train smaller VLMs to pilot the same harness.

On LIBERO-Pro, WAA with skills evolved only from LIBERO-90 reaches a state-of-the-art 75.6% average success, outperforming end-to-end VLAs, code-as-policy agents, and a visual-harness baseline with the same backbone; the same skills remain effective on robosuite without further learning. Fine-tuning Qwen3.5-9B on harness traces raises its out-of-domain success from 1.7% to 43.3%.