Daxia Robot Releases ACE-Brain-0.6 Embodied Foundation Model
Daxia Robot has released ACE-Brain-0.6, an embodied foundation model built on Qwen3.5-9B that is trained on about one million spatial anchoring samples. The company reports that the model outperforms OpenAI GPT-6 Astra in six out of eight Grounding benchmarks. This release represents a notable advancement in spatial perception and vision-language capabilities for robotics. It demonstrates that specialized fine-tuning on open foundation architectures can yield competitive or superior performance in physical AI tasks. ACE-Brain-0.6 achieved an average score of 74.53 across seven public benchmarks compared to 73.68 for GPT-6 Astra, and scored 94.13 in the RobotGrounding real-world scenario evaluation. Daxia Robot was founded by SenseTime co-founder Wang Xiaogang and focuses on physical AI implementation.
## BACKGROUND
Embodied AI refers to AI systems that can interact physically with the world through a robotic body, requiring strong spatial perception and grounding. Grounding in this context means the model's ability to accurately connect language descriptions or commands to specific objects, locations, and coordinates in the physical environment.