Galaxy General launches the world's first general-purpose cerebellum GPT base model for humanoid robots

Galaxy General Robotics has launched AstraBrain-WBC 0.5, the world's first general-purpose cerebellum GPT base model for humanoid robots. It's trained on 20,000 hours of human motion data with 80.4 million parameters, making it the first humanoid robot whole-body real-time motion control model to reach GPT-1 scale.

The model uses a GPT-style causal Transformer architecture, redefining whole-body control as a continuous sequence prediction problem. As training data expanded from 2 million to 2 billion frames, success rate jumped from 83.26% to 92.58%, with zero-shot tracking errors consistently decreasing.

AstraBrain-WBC 0.5 supports coordinated control of dozens of degrees of freedom in milliseconds, helping robots maintain balance, resist disturbances, and complete highly dynamic and precise tasks in complex environments.