9B Model ACE-Brain-0.6 Surpasses GPT-6 Astra in Spatial Grounding Benchmarks
9B模型赢了GPT-6 Astra！大晓机器人新模型专攻空间锚定
Daxiao Robotics has released ACE-Brain-0.6, a visual spatial grounding-enhanced embodied model that outperforms GPT-6 Astra in six out of eight Grounding benchmarks. The model leverages a million high-quality spatial grounding samples, achieving an average score of 74.53 across seven public benchmarks, surpassing GPT-6 Astra's 73.68. It excels in real-world robot scenarios with a score of 94.13 in the RobotGrounding evaluation.
The release of ACE-Brain-0.6 highlights advancements in visual spatial grounding for robotics, demonstrating improved performance over GPT-6 Astra in multiple benchmarks. The model's success is attributed to a large-scale, high-quality dataset and a unified data construction pipeline that integrates diverse sources and annotations.
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.
Reported numbers
training_data_scale
260,000
View original evidence
在保持模型架构与训练方法不变的条件下，团队将训练数据从26万条逐步扩展至百万条，七项公开评测平均得分由69.3分提升至74.5分，呈近似对数线性增长；训练数据规模每翻倍，综合得分平均提升约2.4分。
Open source S20
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
该模型依托约百万条高质量、可验证的空间锚定训练样本，在八项 Grounding 评测中有六项得分超过OpenAI GPT-6 Astra，其中七项公开基准平均得分74.53分，高于GPT-6 Astra的73.68分；在真实机器人场景评测 RobotGrounding 中取得94.13分，超过GPT-6 Astra的91.10分。
Open source S4
在保持模型架构与训练方法不变的条件下，团队将训练数据从26万条逐步扩展至百万条，七项公开评测平均得分由69.3分提升至74.5分，呈近似对数线性增长；训练数据规模每翻倍，综合得分平均提升约2.4分。
Open source S20
大晓机器人整合30余个开源机器人数据家族和私有Ego数据，融合人类第一视角交互、物体与部件密集标注及三维空间数据，将异构数据统一转化为可验证的 空间锚定监督信号 。
Open source S17
Source：Chinese robotics — Dataset and collection discovery · news.qq.com