Xinming Visual Fusion Solution Deploys on Junduo Robotics' Beethoven Dexterous Hand
On October 12, the 2026 China International Industry Fair opened in Shanghai, with industrial embodied intelligence becoming the central focus. Junduo Robotics, a domestic representative in precision execution, drew attention with its newly released "Beethoven" series dexterous hand, whose palm vision solution is exclusively supplied by Xinming. The fair highlighted how embodied AI is moving from demonstration toward industrial application, and dexterous hands are a critical end-effector where perception matters greatly.
When introducing the dexterous hand with 22 reverse-curvature degrees of freedom, Junduo Robotics emphasized its palm vision capability. The hand carries a palm image-depth dual-perception camera, achieving hand-eye integration even as the fingers move through all degrees of freedom. From a palm-sized area comparable to a human hand, it obtains depth and stereo three-dimensional information, enabling real-time object perception, execution, and semantic recognition, and building an integrated perception-execution capability. This means the hand can not only grasp objects but also understand them before acting.
The capability is underpinned by Xinming's dTOF+RGB visual fusion solution. The palm is the most cramped and also the most critical location on a dexterous hand. It leaves only millimeter-level clearance for an optical module, yet the module must simultaneously answer two questions: what is in front of it, and how far away is it. Xinming's answer is to combine both functions in one place, using a spatial intelligence chip to carry depth and color information together.
Xinming's dTOF+RGB visual fusion solution is built on a self-developed spatial intelligence chip. It is currently the only comparable solution on the market that can fit into a palm. Its ultra-small optical architecture raises dTOF resolution to RGB level. Volume is compressed to the extreme, but resolution is not compromised, leaving no weak point in 3D perception. For a space-constrained dexterous hand, this level of integration is a prerequisite for deployment.
First, the solution achieves small size with high resolution. The ultra-small optical architecture pushes dTOF resolution to RGB level, compresses volume to the limit, and keeps 3D perception complete. Conventional depth cameras are too bulky for the palm, so this compactness is essential. Second, it delivers both precision and detail. Millimeter-level ranging provides a reliable distance measurement baseline, while RGB details are fully preserved, keeping contours and textures clear and distinguishable. The system therefore captures both distance and fine detail, balancing near and far information in a single palm module.
Third, the solution offers hardware-level anti-light interference. By suppressing ambient light effects at the source, it remains stable and reliable in complex scenarios such as strong production-line lighting and interlaced shadows. This robustness matters for industrial deployment, where lighting conditions are rarely ideal. Fourth, the design is friendly to timing and computing power. dTOF and RGB are aligned at the hardware level, so multimodal data is naturally synchronized. This removes the need for subsequent alignment computation and significantly reduces the burden on the main controller. The result is a more efficient pipeline from sensing to decision-making.
Fifth, it provides multimodal fusion output. RGB semantics and 3D geometry are unified, and the solution supports multimodal output. This leaves more room for physical AI applications to leverage both appearance and spatial information. Such fused output can support tasks that require recognizing objects while also judging their position and distance. From "seeing" to "seeing accurately," and then to "understanding," Xinming enables a robot's hand to truly see the physical world before it reaches out. Placing spatial intelligence in the palm is the answer Xinming delivers. The combination of compact integration, depth-RGB fusion, and hardware synchronization points to a broader trend: embodied intelligence increasingly depends on perception modules that are small, robust, and semantically aware.