Robot AIBrief
China's Daxiao Robotics says it beats OpenAI's GPT-6 Astra on six of eight spatial tests
Daxiao Robotics has released ACE-Brain-0.6, an AI model that helps robots spot objects and locate themselves in space, Chinese site IT之家 (ITHome) reports on 19 October.
Daxiao Robotics did not start from scratch. ACE-Brain-0.6 is a version of Qwen3.5-9B, retrained with annotated examples. The work is meant to help the robot recognise objects, locate them and know where it can act.
According to the company, the model learned from about one million examples of spatial grounding. It says the model does better than OpenAI's GPT-6 Astra in six of the eight evaluations of this kind.
According to Daxiao's figures, the average score across seven public benchmarks is 74.53 points, against 73.68 for GPT-6 Astra. The RobotGrounding test, which is based on real robot situations, gives 94.13 points against 91.10.
Source
- IT之家 ITHome · ithome.com
.jpg){kind=link}