Daxiao Robotics Releases Embodied Foundation Model ACEBrain 0.6, Outperforms GPT6 Astra in Six of Eight Evaluations
六项评测超越GPT6 Astra，大晓机器人发布具身新模型
Daxiao Robotics has released the embodied foundation model ACEBrain 0.6, which achieved results exceeding international top-tier models like GPT6 Astra in six out of eight spatial grounding evaluations. The model leverages millions of high-quality spatial training data to enhance target recognition, spatial positioning, and interaction point judgment capabilities, advancing robots from understanding the physical world to stable execution of real-world operations.
The release highlights ACEBrain 0.6's performance in spatial grounding tasks, surpassing GPT6 Astra in six out of eight benchmarks, with a focus on data-driven improvements in perception and action alignment.
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
six evaluations exceeded
6
View original evidence
测评结果显示，在八项空间锚定专项评测当中，ACEBrain0.6有六项得分高于GPT6 Astra；七项国际公开基准的平均得分74.53分，高于GPT6 Astra的73.68分；在面向真实机器人场景的自建测试集RobotGrounding拿到94.13分，同样优于Astra的91.10分。
Open source S10
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
这款聚焦视觉空间锚定能力的9B参数模型，凭借百万级高质量空间训练数据，在八项权威评测中有六项成绩超过国际顶尖模型GPT6 Astra，实现国产具身模型专项能力的重要突破。
Open source S3
测评结果显示，在八项空间锚定专项评测当中，ACEBrain0.6有六项得分高于GPT6 Astra；七项国际公开基准的平均得分74.53分，高于GPT6 Astra的73.68分；在面向真实机器人场景的自建测试集RobotGrounding拿到94.13分，同样优于Astra的91.10分。
Open source S10
Source：Chinese robotics — Dataset and collection discovery · sohu.com