The Omnibody Hand glove features tactile sensors, proximity detectors, and global shutter cameras. It captures hand movements and interaction details, including contact, force, and grip patterns.
Reward AI combined visual-inertial systems with electromagnetic positioning. This configuration reduced the average migration error of a target point by 60 percent during high-speed testing.
The OM-1 universal control model processes visual, tactile, and trajectory data to output movement parameters. It learns directly from human demonstrations without requiring telepresence or pre-training on robots.
According to the company, the system learns new tasks, including complex contact sequences, in under 30 minutes of recorded data.
A separate control layer, trained via reinforcement learning in simulation, handles command execution. It operates at high frequency to smooth transitions and compensate for latency or hardware dynamics, enabling cross-platform transfer.
