SimpleAI releases HiFi-UMI: high-fidelity data engine enables direct real-robot deployment without teleoperation data

机器之心 · wechat · 2026-08-04

SimpleAI introduces HiFi-UMI, a high-fidelity data production engine for robot learning. It uses head-mounted stereo inertial SLAM, hardware synchronization, multi-view vision, and full-palm gloves to address issues like trajectory drift, inaccurate bimanual poses, time misalignment, and blind spots in traditional UMI data. Experiments show that policies trained solely on HiFi-UMI data can be deployed on real robots with success rates comparable to teleoperation data. Large-scale pretraining reduces action prediction error by 41% and shows power-law scaling. The team open-sourced the HiFi-UMI-2K dataset.

Original post →

More from Embodied

Embodied channel →