Embodied AI Startup Launches TwinDEX Dexterous Hands for Zero-Robot-Body Data Training
Embodied AI startup Zibianliang has launched TwinDEX, a paired 3-finger, 9-degree-of-freedom (DoF) dexterous hand system consisting of a wearable exoskeleton hand for data collection and a twin hand for robot deployment. The system enables robot model training using purely wearable exoskeleton data without relying on costly teleoperation on physical robot bodies. Data collection is currently the primary bottleneck in embodied AI because real-robot teleoperation requires expensive hardware and dedicated operating environments. TwinDEX boosts data collection efficiency by 5.3 times compared to traditional teleoperation and demonstrates that wearable data can nearly 100% replace physical robot body data during model training. TwinDEX uses a 3-finger design with 7 active DoF, offering a minimal viable solution capable of completing most everyday manipulation tasks with lower hardware complexity. To enable zero-robot-body data transfer, the collection and execution hands are reverse-engineered for high consistency across kinematics, contact mechanics, tactile feedback, and visual appearance.
## BACKGROUND
Embodied AI refers to artificial intelligence systems housed within physical bodies, such as robots, capable of interacting directly with the physical world. Training robots for dexterous manipulation traditionally relies on human operators controlling real physical robots via teleoperation, which severely restricts data scalability due to high costs and facility limitations.