JingShi Technology says its dexterous hand, VIVA, has for the first time played a real-time four-hand piano duet with a person, a milestone it released with CADA, a music-embodied large model that pairs the hand with a score-following brain.
The demo matters because a piano duet is a stiff test for a robot hand. Two players share one keyboard, read the same sheet, and must time every note against a partner they cannot fully predict. For a machine, that means sensing force and position, planning finger moves on the fly, and yielding or leading within milliseconds.
VIVA is built around high-density tactile sensing and fine finger control. JingShi says the hand reports contact and slip continuously and adjusts grip in real time, so it can hold a key with the weight a phrase needs rather than slamming it. CADA supplies the musical side: it tracks the human partner’s tempo and intent and tells the hand which notes to expect and when.
The harder part is the hand-off between the two agents. When the human speeds up or changes dynamics, CADA must re-plan and VIVA must execute without lag or collision. JingShi describes the result as a step toward robots that do not just repeat a script but play with a partner, reading cues the way a human accompanist does.
For the broader humanoid push, the hand is the bottleneck. A body can walk and a brain can plan, but manipulating the physical world still hinges on fingers that feel. Demos like this one show Chinese labs pushing dexterous manipulation from pick-and-place toward tasks with touch, timing and a second agent in the loop.
JingShi positions VIVA as a research and product platform rather than a one-off stunt, with the music case standing in for harder industrial and service jobs that need the same feel-and-respond skill.
Editor’s note: This is an adapted translation of the original OFweek report. It has been trimmed and restructured for readability for an international business audience. The full original (in Chinese) is at https://robot.ofweek.com/2026-09/ART-898890-8120-30704831.html.