Skip to main content

Daxiao's ACE-Brain-0.6 API Gives Robots a Sharper Sense of Space

Daxiao Robotics Opens Its Spatial Intelligence API to Developers

On October 29, Daxiao Robotics released ACE-Brain-0.6, an embodied spatial intelligence API that hands developers a set of "spatial perception nerves" for machines. The interface tackles three core tasks: recognizing what object is in front of a robot, pinpointing that object in 3D space, and figuring out where people and things can interact. In plain terms, it gives machines a working sense of the physical world around them.

What Made the Brain Smarter

The upgrade comes down to data. Daxiao fed the model roughly 1.5 million spatial anchoring training entries, up from about 1.1 million—a nearly 40% boost to its spatial memory. With that thicker foundation, ACE-Brain-0.6 climbed to an average score of 75.41 across seven public Grounding benchmarks. That not only holds above its previous level but also edges past OpenAI's GPT-6 Astra.

Beating a frontier flagship on the foundational track of spatial understanding is a meaningful step for a domestic model aimed at embodied intelligence. It brings robots closer to the basic promise of seeing clearly and moving to the right places—without constant human correction.

Why It Matters for Developers

For teams building robotics and embodied AI applications, the API removes a major hurdle: they no longer have to train spatial reasoning from scratch. Instead, they can plug in a ready-made capability that already understands targets, locations, and interaction zones. That could speed up everything from warehouse automation to service robots that navigate crowded rooms.

Daxiao's move also signals a broader trend: spatial intelligence is becoming a platform service rather than a proprietary research project. As more developers get access, expect faster iteration on robots that can actually make sense of their surroundings.

Key Points

  • ACE-Brain-0.6 API launched October 29, offering spatial recognition, 3D localization, and interaction-zone detection.
  • Training data grew from 1.1 million to 1.5 million spatial anchoring entries, a ~40% increase.
  • Average score of 75.41 across seven public Grounding benchmarks, surpassing GPT-6 Astra.
  • The API gives developers a plug-and-play spatial intelligence layer for embodied AI applications.