# SO-101 Bench发布：仿真与真机策略排名完全一致

> 原文链接：https://robotics.hellobike.com/papers/2026-10-27-2610-36420-41c37558
> 频道：papers | 更新日期：2026-10-27 | 标签：auto, arxiv, hf-daily-papers
> 核心摘要：面向低成本机械臂SO-101推出Isaac Lab仿真基准，含4类任务与3629条演示。验证显示5种策略在虚实环境排名Spearman相关系数达1.00，LLM代码生成代理以83.2%成功率显著优于纯VLA模型。

# SO-101 Bench发布：仿真与真机策略排名完全一致

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[SO-101 Bench: Accessible Spatial Manipulation](https://arxiv.org/abs/2610.36420)（HF Daily Papers · arXiv 2610.36420）

## 核心内容

针对低成本机械臂SO-101缺乏开箱即用空间操作策略及现有基准依赖真机、任务单一的问题，研究团队发布SO-101 Bench仿真基准。该基准基于真实SO-101 setups构建Isaac Lab数字孪生，包含4类任务5个评分族、138种物体（涵盖可见/未见实例与类别）及916个评估回合。所有学习策略均在3629条模拟演示上以同等算力微调，旨在解决具身智能研究中Sim-to-Real评估不一致的核心痛点。

实测表明，5种无真数据训练的策略（含4种VLA和1个LLM Agent）在真机未见环境中保持与仿真完全一致的排名（Spearman rho=1.00, MMRV=0.000）。9种VLA与世界模型最优成功率约61%，而LLM Agent通过代码生成调用底层抓取策略，成功率达83.2%。消融实验证实该优势源于语言模型选择与执行框架设计，而非单纯感知能力，揭示了当前VLA在旋转适配规划与遮挡记忆等空间推理任务上的局限。

## 关键看点

- **虚实一致性**：5种策略在仿真与真机未见环境排名Spearman相关系数达1.00，MMRV为0.000，验证了Isaac Lab数字孪生的高保真度。
- **LLM代理优势**：LLM Agent通过代码生成调度抓取策略，成功率83.2%，比最优纯VLA模型高出约22个百分点，凸显程序化推理价值。
- **数据规模**：提供3629条标准化模拟演示与916个评估回合，覆盖138种物体的可见/未见类别分割，统一了低成本臂算法评测基线。
- **能力瓶颈**：最优学习策略成功率仅61%，空间精确放置、未见物体泛化及旋转适配规划仍是开放难题，指明后续算法优化方向。

## 工程与产业观察

该基准首次量化验证了低成本机械臂仿真与真机策略排名的完美对齐，为具身算法提供了可信的零样本迁移评估标尺。LLM Agent大幅超越端到端VLA的事实表明，现阶段将高层语义推理与底层控制解耦，利用代码作为中间表征处理复杂空间约束，仍是比纯神经网络更高效的工程路径。

## 英文摘要 (Abstract)

Low-cost arms such as the SO-101 have become the most common entry point to robot learning, yet no released policy performs spatially constrained, language-conditioned tasks on them out of the box, and existing SO-101 benchmarks run only on physical hardware with simple tasks and few held-out objects. We introduce SO-101 Bench, a simulated benchmark for language-conditioned spatial manipulation built on an Isaac Lab digital twin of a real SO-101 setup. It has four tasks scored as five families, 138 objects in seen, unseen-instance and unseen-class splits, 916 evaluation episodes with subsets for held-out appearances, rotate-to-fit planning and occlusion memory, and 3,629 simulated demonstrations on which every learned policy is fine-tuned under an equal compute budget. The twin ranks policies as the real robot does: five policies trained without real data, four VLAs and an LLM agent, keep their exact ranking when the same weights run on the physical arm in an unseen environment with unseen objects (Spearman rho = 1.00, MMRV = 0.000), although simulation overestimates success. Across nine VLAs and world-action models, the best learned policies score near 61%, with spatial placement, unseen object classes and rotate-to-fit planning remaining open. An LLM agent that writes code and hands grasping to a learned policy scores 83.2%, and an ablation shows that this lead comes from the choice of harness and the choice of language model.

Hugging Face Daily Papers 热度：4 upvotes
作者：Truman Hickok
:::

## 原始参考来源 (Sources)

- [HF Daily Papers · arXiv 2610.36420：SO-101 Bench: Accessible Spatial Manipulation](https://arxiv.org/abs/2610.36420)