# 通过语义与物理双重纠正使用视觉-语言模型重规划人机协作任务

> 原文链接：https://robotics.hellobike.com/papers/2026-10-01-2602-14551
> 频道：papers | 更新日期：2026-10-01 | 标签：arxiv, cs.RO
> 核心摘要：面向人机协作装配中的模糊纠正指令，提出结合内外纠正、6-DoF抓取与无碰撞规划的VLM重规划框架，并在仿真和真机验证。

# 通过语义与物理双重纠正使用视觉-语言模型重规划人机协作任务

::: v-pre
> 原文链接：[Replanning human–robot collaborative tasks with vision–language models via semantic and physical dual–correction](https://arxiv.org/abs/2602.14551)（arXiv 2602.14551 · 发布日期：2026-10-01）

**Replanning human–robot collaborative tasks with vision–language models via semantic and physical dual–correction**

面向人机协作装配中的模糊纠正指令，提出结合内外纠正、6-DoF抓取与无碰撞规划的VLM重规划框架，并在仿真和真机验证。

## 核心要点

- 人机协作装配要求机器人理解模糊纠正指令，同时生成物理可执行动作。
- 框架将指令映射为含抓取位置和工具选择的动作目标候选，并用执行前内部纠正与执行后外部纠正验证。
- 系统结合VLM推理、6-DoF抓取生成和无碰撞轨迹规划；仿真消融显示内外纠正效果依配置而变。
- 真机上物体固定任务抓取位置调整成功率66.7%，工具准备任务初选100%、纠正选择75.0%。

**验证范围与局限**：验证包括仿真消融和一台上半身人形机器人上的物体固定、工具准备任务，视觉状态验证是假阴性导致成功率下降的关键限制。

## 英文摘要 (Abstract)

Human–robot collaborative assembly requires robots to interpret ambiguous corrective instructions while producing physically executable motions. Vision–language models (VLMs) provide semantic reasoning but may select logically inconsistent targets or misjudge execution outcomes. We propose a replanning framework that maps human instructions to Action Target candidates, including grasp position and tool selections, and combines an Internal Correction Model for pre-execution logical verification with an External Correction Model for post-execution visual verification. The framework integrates VLM reasoning with 6-DoF grasp generation and collision-free trajectory planning. Simulation ablations show configuration-dependent effects: internal correction improves candidate validity, whereas external correction enables recovery for a low-latency VLM but can reduce success when visual verification produces false negatives. Real-world experiments with an upper-body humanoid robot achieved a 66.7% success rate for grasp position adjustment in the object fixation task; in the tool preparation task, success rates were 100% for initial selection and 75.0% for corrective selection. These results demonstrate interactive replanning across physical and semantic collaborative tasks while identifying visual-state verification as a key limitation.

作者：Taichi Kato, Takuya Kiyokawa, Namiko Saito, Kensuke Harada
:::

## 原始参考来源 (Sources)

- [arXiv 2602.14551：Replanning human–robot collaborative tasks with vision–language models via semantic and physical dual–correction](https://arxiv.org/abs/2602.14551)