ACE Robotics releases spatial perception AI model for robotics

ACE Robotics has introduced ACE-Brain-0.6, a new embodied AI model developed through supervised fine-tuning of the Qwen3.5-9B architecture. The technology is specifically designed to assist robots in identifying objects, calculating their precise spatial coordinates, and determining viable interaction points within a physical environment.

The firm states that the model was trained on a dataset of approximately one million spatial-grounding samples. In internal and public benchmarks, ACE-Brain-0.6 reportedly outperformed GPT-6 Astra on six of eight metrics focused on spatial perception.

In the RobotGrounding evaluation, which simulates real-world robot scenarios, ACE-Brain-0.6 scored 94.13, while Astra recorded 91.10. These results suggest a growing focus on integrating high-level visual-spatial reasoning into robotics to improve autonomous task execution.