ACE-Brain-0.6 Model Surpasses GPT-6 Astra in Visual-Spatial Anchoring Evaluations
9B模型跑赢GPT-6 Astra，大晓机器人用一百万条数据回答了一个行业问题
Daxiao Robotics achieved six leading positions out of eight visual-spatial anchoring evaluations using the ACE-Brain-0.6 model, with a score of 94.13 on the RobotGrounding evaluation set, approximately 3 points higher than GPT-6 Astra. The article analyzes the model's performance in tasks such as spatial perception and target missing recognition, and discusses the key role of data diversity in enhancing model performance.
It also outlines the methodology and pipeline for generating annotated robot interaction data.
Source: Chinese robotics — Dataset and collection discovery · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
此次对比中有一项数据来自大晓自建的RobotGrounding评测集：2591道题，覆盖目标定位、小物体识别、多实例计数、目标缺失判断、跨相机视角对应等14类任务。
Open source S10
在这套更贴近真实机器人作业场景的题目上，ACE-Brain-0.6取得94.13分，领先Astra约3.03分，也高于参与对比的Google Gemini 4 Argon、Anthropic Claude Opus 5.5，以及达摩院、智源研究院等国内外代表性模型。
Open source S11
在模型架构与训练方法完全不变的条件下，团队将训练数据从26万条逐步扩展至约百万条，七项公开基准的平均得分从69.3分升至74.5分，呈现近似对数线性的增长——数据规模每翻倍，综合得分平均提升约2.4分。
Open source S12
Source：Chinese robotics — Dataset and collection discovery · news.qq.com