High-quality human demonstration data is becoming an essential foundation for embodied AI, imitation learning, and robotic manipulation research. However, recording a person completing a task is only the beginning. Research teams also need accurately aligned information about hand movement, gripper status, visual context, depth, motion, and spatial trajectories.

When these data streams are captured by separate devices, differences in timing, coordinate systems, and data quality can make the resulting dataset difficult to use. Problems may only become visible after collection is complete, leading to repeated demonstrations and additional post-processing work.