外观
论文HF Daily Papers · arXiv 2610.270409 天前
ACE-Brain-0.6发布：150万样本微调9B模型空间感知超GPT-6 Astra
本文由 AI 根据公开内容整理，可能有误，请以原文为准：ACE-Brain-0.6: Scaling Spatial Grounding for Physical Foundation Models（HF Daily Papers · arXiv 2610.27040）
核心内容
针对通用多模态模型在具身操作中空间感知不足的问题，研究团队推出ACE-Brain-0.6物理基础模型。该工作核心在于构建了统一数据管线，将机器人操作录像与多源互补视觉数据融合，并自动转化为可验证的空间接地（Grounding）监督信号。使用约150万条此类样本对Qwen3.5-9B进行微调，使模型在7个公共接地基准测试中的平均得分达到75.41，其中4项指标超越GPT-6 Astra（73.68），验证了通过数据缩放增强开源9B模型具身感知能力的有效性。
技术路径上，ACE-Brain-0.6强调“可验证接地监督”的数据质量而非单纯堆量。实验表明，在固定Qwen3.5-9B架构与微调配置下，随着训练集覆盖更多样化的接地场景，模型聚合性能呈稳步上升趋势。这揭示了具身智能Scaling Law的新维度：相较于纯文本或通用图像数据，高质量、带物理交互真值的结构化接地数据，是提升小参数量模型从高层推理到物理场景映射能力的关键瓶颈突破口，为端侧具身模型部署提供了高性价比范式。
关键看点
- 性能对标：ACE-Brain-0.6在7项公共接地基准中均分75.41，4项超越GPT-6 Astra的73.68，确立9B级开源模型空间感知新SOTA
- 数据规模：构建统一管线融合机器人操作与多源视觉数据，生成约150万条可验证空间接地监督样本用于模型微调
- 基座架构：基于Qwen3.5-9B进行微调，证明在固定算力与架构下，扩展多样化接地场景数据可持续提升模型性能
- 核心机制：将机器人录像自动转换为可验证接地监督信号，解决高层语义推理与底层物理交互场景的对齐难题
工程与产业观察
该研究为具身智能数据工程提供关键实证：在模型参数量受限（9B）时，构建“可验证空间接地”数据集比通用预训练更能提升操作性能。对于端侧部署而言，这意味着可通过清洗机器人日志与合成数据来弥补算力短板，降低对闭源大模型的依赖，是低成本实现高精度具身感知的可行路径。
英文摘要 (Abstract)
General-purpose multimodal models, including GPT-6 Astra and Claude Opus 5.5, demonstrate broad capabilities in embodied manipulation. Our analysis indicates that this generality is accompanied by strong spatial perception, as Astra can determine where to interact with task-relevant entities amid visual distractors. Because spatial perception connects high-level reasoning to the physical scene, we investigate scaling visual grounding to strengthen the perceptual foundation of general embodied operation. We develop a unified data pipeline that combines robot manipulation recordings with complementary visual data from diverse sources. After automatically annotating the robot recordings, the pipeline converts the collected data into verifiable grounding supervision. Fine-tuning Qwen3.5-9B on approximately 1.5 million samples produced by this pipeline yields ACE-Brain-0.6. The model outperforms GPT-6 Astra on four of seven public grounding benchmarks, with an overall average score of 75.41 compared with Astra's 73.68. Under a fixed Qwen3.5-9B fine-tuning setup, aggregate grounding performance improves steadily as successive data releases expand the training collection to cover a broader range of grounding scenarios. These results show that scaling verifiable grounding supervision from diverse sources can strengthen the spatial perception of an open 9B model for general embodied operation.
Hugging Face Daily Papers 热度：5 upvotes 作者：Yinhe Wang, Tao Tao, Zexin He, Shuang Luo, Wei Mao, Yao Mu, Feng Pan, Hanming Deng, Lewei Lu, Zhi Hou, Xingang Pan, Dahua Lin, Dacheng Tao, Xiaogang Wang
来源
本页事实对应的公开页面，以原文为准。