HiFi-UMI: Learning Deployable Manipulation Policies from High-Fidelity UMI Data Alone

HiFi-UMI presents a robot-free data-capture system designed to make UMI demonstrations accurate enough for direct real-robot policy deployment. It combines head-mounted offline stereo-inertial SLAM, native inter-gripper relative pose, microsecond GPIO synchronization, and two wide-angle cameras per hand covering about 200 degrees. The system reports 3 mm workspace-local end-effector accuracy without external tracking. Policies post-trained only on HiFi-UMI data matched teleoperation performance across StarVLA-QwenPI, OpenPI-pi_0.5, and LingBot-VA, with success-rate differences of -2.5, +3.1, and -0.6 percentage points. The released HiFi-UMI-2K corpus contains 2,000 hours of demonstrations.