Paired wearable and laboratory recordings provide the foundation for developing the Human Motion Foundation Model. Data collection across multiple sites, supported by refinement and quality checks, is intended to capture a wider range of people and movement conditions. Synthetic IMU signals will complement measured recordings.

The resulting model is intended to support human movement research and wearable assistance by connecting wearable measurements with full-body motion and force estimates. These outputs provide a basis for applications that account for how people move.

Synthetic IMU generation converts recorded human motion into acceleration and gyroscope signals. Motion-capture recordings provide the basis for physics-based signal calculation. Neural models then refine these signals to better match real IMU recordings.

This approach combines physics-based signal generation with learning from measured data. The generated signals are intended to complement recorded IMU data and support training of the Human Motion Foundation Model to recover full-body movement and biomechanical loading from sparse wearable measurements.

The proposed approach uses a motion generator to provide a reference trajectory. An imitation policy learns to control a simulated body to follow that trajectory inside a physics simulator. Contact with the ground, gravity and joint limits constrain how the body moves.

This connects generated motion with physically plausible execution. Alongside the resulting movement, the simulator provides estimates of ground-reaction forces, joint moments and power. These outputs allow researchers to examine both how the body moves and the mechanical demands involved in performing the movement.