Skip to content
IT Home — Robotics·· 4 hours agoEditorial score63

Da Xiao Robot Releases Embodied Base Model ACE-Brain-0.6, Surpasses OpenAI GPT-6 Astra in Six Spatial Perception Benchmarks

大晓机器人发布具身基模型 ACE-Brain-0.6，六项空间感知评测超越 OpenAI GPT-6 Astra

Summary

Da Xiao Robot has released the visual spatial anchoring enhanced embodied base model ACE-Brain-0.6, which was fine-tuned on Qwen3.5-9B. The model outperformed OpenAI's GPT-6 Astra in six out of eight spatial grounding benchmarks, achieving an average score of 74.53 compared to 73.68, and scored 94.13 in the RobotGrounding benchmark, surpassing Astra's 91.10.

Source: IT Home — Robotics · 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

  • average score in public benchmarks

    74.53

    View original evidence
    其中，七项公开基准平均得分为 74.53 分，高于后者的 73.68 分；在真实机器人场景评测 RobotGrounding 中取得 94.13 分，超越 Astra 的 91.10 分。
    Open source S4

  • score in RobotGrounding evaluations

    94.13

    View original evidence
    其中，七项公开基准平均得分为 74.53 分，高于后者的 73.68 分；在真实机器人场景评测 RobotGrounding 中取得 94.13 分，超越 Astra 的 91.10 分。
    Open source S4

  • score of GPT-6 Astra in public benchmarks

    73.68

    View original evidence
    其中，七项公开基准平均得分为 74.53 分，高于后者的 73.68 分；在真实机器人场景评测 RobotGrounding 中取得 94.13 分，超越 Astra 的 91.10 分。
    Open source S4

  • score of GPT-6 Astra in RobotGrounding evaluations

    91.1

    View original evidence
    其中，七项公开基准平均得分为 74.53 分，高于后者的 73.68 分；在真实机器人场景评测 RobotGrounding 中取得 94.13 分，超越 Astra 的 91.10 分。
    Open source S4

  • number of Grounding evaluations with higher scores

    6

    View original evidence
    据大晓披露，该模型使用约百万条空间锚定训练样本，在八项 Grounding 评测中有六项成绩超过 OpenAI GPT-6 Astra。
    Open source S3
  • paper: not_reported
  • code: not_reported
  • weights: not_reported
  • dataset: not_reported
Source excerpts and review record

Automatically extracted; no manual editorial approval recorded.

据大晓披露，该模型使用约百万条空间锚定训练样本，在八项 Grounding 评测中有六项成绩超过 OpenAI GPT-6 Astra。

Open source S3

其中，七项公开基准平均得分为 74.53 分，高于后者的 73.68 分；在真实机器人场景评测 RobotGrounding 中取得 94.13 分，超越 Astra 的 91.10 分。

Open source S4

Source：IT Home — Robotics · ithome.com

#Model release#Evaluation#Robot policies#Imitation learning#World models
Timezone · UTC

Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).