Autonomous Agent Developer Unveils TwinDEX, a Twin Dexterous Hand Achieving True Robot-Free Training with Comparable Results
On September 2, 2026, the embodied robotics company Independent Variables officially launched TwinDEX, a three-finger, nine-degree-of-freedom dexterous manipulation hand. The product forms a twin system combining a wearable capture unit and a robot deployment unit, achieving for the first time purely robot-free data-driven dexterous manipulation training. Evaluations show that its robot-free data collection efficiency reaches 5.3 times that of real-robot teleoperation, with training outcomes nearly 100% equivalent to the real-hardware approach. The development breaks through the industry’s long-standing “triple challenge” of data quality, dexterity, and scalability, delivering a low-cost, high-efficiency data infrastructure for embodied intelligence.
In response to bottlenecks in dexterous manipulation data acquisition, Independent Variables has overcome three core technical barriers: high-precision dexterous tracking of the human hand, cross-embodiment kinematic mapping, and sim-to-real data migration. TwinDEX enables high-frequency, high-precision capture of human hand movements and converts them into robot control instructions through an advanced mapping algorithm, allowing flexible manipulation across multiple robotic platforms without prior hardware-specific data collection. A senior engineer at the company noted that while real-robot teleoperation remains a conventional data source for dexterous tasks, it is severely constrained by cost, time, and scenario adaptability. In contrast, TwinDEX turns everyday operator actions into usable training data, fundamentally expanding the limits of data scale and diversity.
For practical training, TwinDEX offers two distinct application modes. When real-robot execution is needed, the system integrates with existing robotic arms to fine-tune end-effector control and closed-loop compensation in hardware. When only training is required, it operates purely in simulation, using the captured high-precision motion data to pre-train policy models that are then directly transferred to real-world robots. This dual-mode design significantly lowers the entry threshold for embodied intelligence development, enabling rapid iteration from data collection to deployment.
It is reported that TwinDEX supports multi-brand robotic arm configurations, including UFactory and AgileX, and provides a complete SDK and API interface to facilitate secondary development. The company stated that it will continue to deepen the twin system architecture, advance dexterous manipulation into broader service and industrial scenarios, and forge an open ecosystem for embodied data infrastructure. Independent Variables is a dedicated innovator in embodied robotics in China, focusing on generic dexterous operation and efficient data systems, with its research findings applied across manufacturing, logistics, and home services.
