NVIDIA research team has announced MotionBricks, a motion generation framework for game animation and robot control, accepted as a SIGGRAPH 2026 paper by ACM Transactions on Graphics.
MotionBricks handles over 350,000 motion skills on a single neural backbone, achieving a throughput of 15,000 FPS with 2 millisecond latency. The framework was developed by researchers from NVIDIA, ETH Zurich, Simon Fraser University, and the University of Texas at Austin.
The main dataset consists of approximately 700 hours of high-quality motion capture, comprising 350,000 motion clips, 9,300 different skills, 36 categories, and over 163 performers.
MotionBricks uses 'smart primitives' to handle the complexity of motion states without retraining or dedicated tagging for each downstream task. At the higher level, 'smart locomotion' creates natural movement from arbitrary speed, direction, and style specifications, capable of generating styles like injured walking, zombie-like movements, skipping, and strafe from a single prompt in zero shots.
Critically, the framework has been validated on a Unitree G1 humanoid robot for full-body control in real-world environments, bridging the gap between virtual character animation and physical robot control. The research team says one of its key contributions is enabling unified motion synthesis across both domains.
An early preview version of the code is available on GitHub, including a lightweight MotionBricks-controlled Unitree G1 demo. The team aims for a full release around July 2026, including a model fully integrated into the GR00T Whole-Body Control robot control framework.
