Wetour Robotics demonstrates dual-modal sEMG-vision system to improve humanoid robot training data capture

Wetour Robotics released a development demo of its Orchestra system combining sEMG wrist sensing with first-person vision for robot training data.

• Design targets vision-only gaps in force measurement, addressing occlusion; internal test logged 21.8% hand-tracking frame failures, longest dropout 4.32 seconds.
• On-device model outputs 20 hand joint angles; benchmark showed 50.4 milliseconds to process a 1-second sEMG window with zero lookahead.
• Force-estimation pipeline aims to convert sEMG into kilogram-equivalent grasp force; validated end-to-end on synthetic data, live wristband validation still in development.
• System under development to tag reach, grasp, hold, release events to produce structured demonstrations for imitation learning.