# ForceHarness框架赋能冻结模型零样本力控，GPT-6 Astra成功率升至90%

> 原文链接：https://robotics.hellobike.com/papers/2026-11-16-2611-21099-90b832d9
> 频道：papers | 更新日期：2026-11-16 | 标签：auto, arxiv, hf-daily-papers
> 核心摘要：针对前沿多模态模型无法感知接触力的问题，ForceHarness通过自终止力原语与物理翻译层实现零样本力控。实测使GPT-6 Astra在无机腕传感器下成功率从3.3%跃升至90%，仿真性能超越示教策略并匹配专家控制器。

# ForceHarness框架赋能冻结模型零样本力控，GPT-6 Astra成功率升至90%

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[ForceHarness: Turning Force into Action and Physics for Frozen Frontier Models](https://arxiv.org/abs/2611.21099)（HF Daily Papers · arXiv 2611.21099）

## 核心内容

针对冻结前沿模型难以处理高频接触力及原始力数据语义缺失的难题，研究提出ForceHarness框架。该方案无需微调或演示即可让API模型具备力控能力。在真实机械臂无腕部力传感器条件下，GPT-6 Astra的零样本操作成功率由3.3%提升至90.0%；在MuJoCo仿真中，HiddenPhys基准得分提升34.3个百分点，ForceBench提升17.5个百分点，性能全面超越基于示教训练的力感知策略。

技术架构包含两大核心：一是自终止力原语（Self-terminating force primitives），在模型决策回路之下自主调节力并在接触事件时返回，解决API延迟与接触演化速度不匹配问题；二是物理翻译层，将力轨迹转化为带不确定性的物理量作为工具调用结果，替代模型无法理解的原始力流。配套构建的HiddenPhys七任务基准包含隐藏物理变量，消融实验证实小参数模型在接收结构化物理量后决策质量显著优于直接输入原始力值。

## 关键看点

- **零样本力控突破**：ForceHarness使GPT-6 Astra在无腕部力传感器的真机环境中，零样本操作成功率从3.3%大幅提升至90.0%
- **底层力控解耦**：采用自终止力原语在模型API决策循环之下自主闭环调节接触力，解决视觉语言模型推理延迟与高频力控时序冲突
- **语义化物理反馈**：物理翻译层将原始力轨迹转换为带不确定性的物理量，使冻结模型能以工具调用形式理解力学状态而非处理原始信号
- **仿真性能验证**：在HiddenPhys与ForceBench基准上分别提升34.3和17.5个百分点，超越示教训练策略并匹配手写专家控制器水平

## 工程与产业观察

该工作证明了在不开放模型权重前提下，通过系统级中间件弥补具身感知短板的有效性。将高频力控下沉为原子技能、将传感数据抽象为语义化物理量，是解决通用大模型落地机器人“最后一公里”的关键工程范式，为非侵入式增强冻结模型物理交互能力提供了标准化参考路径。

## 英文摘要 (Abstract)

Frontier multimodal models can now control robots zero-shot from images, yet many manipulation tasks depend on contact forces that images cannot show. Bringing force to an API-only frozen model raises two challenges. First, contact evolves much faster than an API model can decide, so the system must close the loop on force beneath the model. Second, a frozen model cannot interpret raw force streams, so force must be returned in a form the model can reason about. Therefore, we present ForceHarness, a harness that lets frozen models use force without demonstrations or finetuning. After calibrating hand force from a wrist sensor or joint torques alone, ForceHarness addresses the first challenge with self-terminating force primitives that regulate force on their own and return at a contact event. To address the second challenge, a physics translation layer turns each force trace into physical quantities with uncertainty as the tool result. To evaluate ForceHarness against expert force controllers and learned policies under identical conditions, we also build HiddenPhys, a seven-task MuJoCo benchmark with hidden physical variables. On a real arm without a wrist force sensor, ForceHarness raises the zero-shot success rate of GPT-6 Astra from 3.3% to 90.0%. In simulation, it adds 17.5 percentage points on the public ForceBench suite and 34.3 percentage points on HiddenPhys, surpassing force-aware policies trained on demonstrations on both benchmarks and matching hand-written expert controllers on HiddenPhys. Ablations further show that smaller models make significantly better decisions when given physical quantities rather than raw force values.

Hugging Face Daily Papers 热度：19 upvotes
作者：Ke Sun, Kevin Yuchen Ma, Yike Zhang, Yanzhe Chen, Mike Zheng Shou
:::

## 原始参考来源 (Sources)

- [HF Daily Papers · arXiv 2611.21099：ForceHarness: Turning Force into Action and Physics for Frozen Frontier Models](https://arxiv.org/abs/2611.21099)