X Square Robot

X Square Robot's TwinDEX is a hardware plus data system aimed at teams training dexterous manipulation policies without spending weeks teleoperating robots. The product pairs a wearable, three-finger, nine-degree-of-freedom exoskeleton for data collection with a closely matched robotic end effector for deployment, sharing kinematic chains, joint axes, link proportions, contact geometry, surface materials, visual appearance, and sensor placement so measured finger states map directly to robot joint space without complex retargeting. Robots operating in open environments need to twist, insert, pour, open latches, use tools, handle narrow spaces, and adjust through contact, and training data for those closed-loop behaviors remains a major bottleneck because on-robot teleoperation occupies the robot, requires a prepared workspace, and often depends on a skilled operator. The launch matters because it puts a concrete number on the wearable-collection approach. In evaluations across tasks like cap twisting, sweeping with a broom and dustpan, releasing toolbox latches, and operating a syringe, TwinDEX delivered up to 5.3 times the effective throughput of on-robot teleoperation. The team then ran a full standardized chemistry experiment autonomously in a single uncut run across 24 sub-actions, including opening containers, handling a thin scooper, using a rubber-bulb pipette, holding a nearly transparent glass rod against a beaker, guiding a pour, and switching tools, with the policy learning this long-horizon behavior from only a few hundred robot-free episodes and no on-robot training data. X Square Robot is a Shenzhen-based embodied AI company led by founder and CEO Wang Qian, with co-founder Wang Hao serving as CTO, and it has focused since day one on in-house development of foundation models for robotics. For operator-facing teams evaluating physical AI stacks, TwinDEX is worth a look as a self-contained answer to the demonstration-data problem, letting one person, one table, and one wearable rig produce training data that transfers to a matching robot arm.