Giving Robot Hands a Sense of Touch: DAIMON Robotics' Tactile Revolution

Imagine robots that can feel the texture of a fabric, sense the force needed to pick up a delicate object, or adjust their grip in real time. This is the vision of DAIMON Robotics, a Hong Kong-based company that has just released the largest open-source dataset for robotic touch. By combining high-resolution tactile sensors with a new AI architecture, they aim to give robot hands the sensitivity that has long been missing. Below, we explore the key questions about this breakthrough and what it means for the future of physical AI. <h2 id="question1">What is the Daimon-Infinity dataset, and why is it important?</h2> Daimon-Infinity is the world's largest omni-modal robotic dataset for physical AI, announced by DAIMON Robotics in April 2024. It includes high-resolution tactile sensing data from over 80 real-world scenarios and 2,000+ human skills, spanning tasks from folding laundry at home to precision assembly in factories. With millions of hours of multimodal data — including vision, tactile, language, and action — it is designed to train robots to interact with the physical world more deftly. The dataset was created in collaboration with partners like Google DeepMind, Northwestern University, and the National University of Singapore. Its importance lies in addressing a critical gap: most robot training relies on vision and language, ignoring the sense of touch. By making tactile data available at scale, Daimon-Infinity can accelerate the development of robots that can feel and adapt, moving beyond rigid, pre-programmed behaviors. The project also open-sources 10,000 hours of its data to spur global research.<figure style="margin:20px 0"><img src="https://spectrum.ieee.org/media-library/man-wearing-glasses-and-a-gray-shirt-smiles-at-camera-while-surrounded-by-futuristic-robots-and-tech-devices-in-a-photo-illustra.jpg?id=66444415&width=980" alt="Giving Robot Hands a Sense of Touch: DAIMON Robotics' Tactile Revolution" style="width:100%;height:auto;border-radius:8px" loading="lazy"><figcaption style="font-size:12px;color:#666;margin-top:5px">Source: spectrum.ieee.org</figcaption></figure> <h2 id="question2">Why release a dataset instead of focusing solely on products?</h2> DAIMON Robotics believes that data is the fuel for embodied AI. While the company has advanced tactile sensor hardware — a fingertip-sized module with over 110,000 sensing units — they recognized that even the best sensors are useless without rich, diverse training data. By releasing Daimon-Infinity now, they aim to create a community standard that researchers and companies worldwide can use to develop and benchmark tactile-aware algorithms. This accelerates the entire field, rather than keeping the advantage solely for themselves. Co-founder Prof. Michael Yu Wang explains that collaboration with academic institutions like HKUST and industry partners like Google DeepMind ensures the dataset covers a wide range of tasks and environments, from cluttered homes to structured factories. The open-source component lowers the barrier for startups and labs, fostering innovation that could lead to faster deployment of touch-enabled robots in real-world settings such as hotels and convenience stores in China. <h2 id="question3">How does tactile feedback improve robot manipulation?</h2> Current robot systems rely heavily on the Vision-Language-Action (VLA) model, which uses cameras and language prompts to control motion. However, this approach is blind to touch — a robot may see an object but not know whether it is slipping from its grip, whether the surface is too hard, or how much force to apply. Tactile feedback fills this gap by providing real-time data on pressure, texture, and shear forces. For example, when picking up a ripe tomato, visual input alone cannot determine the optimal grip force; tactile sensors can detect slight deformations and adjust accordingly. Similarly, in assembly tasks, a robot can feel when a part is seated properly. Prof. Wang’s team has shown that integrating tactile data reduces errors in dexterous manipulation tasks by up to 40%. The result is robots that can handle fragile objects, perform fine motor skills like threading a needle, and adapt to unpredictable environments — capabilities essential for robots working alongside humans in homes, hospitals, and factories. <h2 id="question4">What is the Vision-Tactile-Language-Action (VTLA) architecture?</h2> The VTLA architecture, pioneered by Prof. Michael Yu Wang and his team at DAIMON Robotics, stands for Vision-Tactile-Language-Action. It elevates tactile sensing to a first-class modality, equal to vision and language, in the AI model that controls robot manipulation. In the standard VLA model, vision and language input are processed to generate action commands, but touch is ignored or treated as secondary. VTLA explicitly incorporates tactile data as a separate, parallel input channel. This means the robot’s AI can learn correlations between what it sees, what it feels, and the actions it needs to take. For instance, when folding a towel, VTLA allows the robot to not only see the folds but also feel the texture and tension of the fabric, enabling more precise folding. The architecture is trained on the Daimon-Infinity dataset, which includes synchronized vision, tactile, language, and action recordings from thousands of human demonstrations. This holistic approach promises robots that can understand and interact with the physical world more intuitively, much like humans do. <h2 id="question5">Who is Prof. Michael Yu Wang, and what is his background?</h2> Prof. Michael Yu Wang is the co-founder and chief scientist of DAIMON Robotics. He earned his PhD in robotics at Carnegie Mellon University under the supervision of Matt Mason, a pioneer in robot manipulation. He later moved to Hong Kong, where he founded the Robotics Institute at the Hong Kong University of Science and Technology (HKUST). An IEEE Fellow and former Editor-in-Chief of IEEE Transactions on Automation Science and Engineering, Prof. Wang has spent nearly four decades advancing the field. His research focused on dexterous manipulation, and he identified a critical problem: even state-of-the-art robots lacked tactile sensitivity. This insight led him to develop the Vision-Tactile-Language-Action (VTLA) architecture and to guide DAIMON in creating high-resolution tactile sensors and the Daimon-Infinity dataset. His goal is to make robot manipulation as touch-aware as human handling, enabling robots to work safely and effectively in unstructured, natural environments. He envisions the dataset as a catalyst for a new generation of physical AI.<figure style="margin:20px 0"><img src="https://spectrum.ieee.org/media-library/man-wearing-glasses-and-a-gray-shirt-smiles-at-camera-while-surrounded-by-futuristic-robots-and-tech-devices-in-a-photo-illustra.jpg?id=66444415&width=1200&height=600&coordinates=0%2C123%2C0%2C123" alt="Giving Robot Hands a Sense of Touch: DAIMON Robotics' Tactile Revolution" style="width:100%;height:auto;border-radius:8px" loading="lazy"><figcaption style="font-size:12px;color:#666;margin-top:5px">Source: spectrum.ieee.org</figcaption></figure> <h2 id="question6">Where will touch-enabled robots first be deployed in the real world?</h2> According to Prof. Wang, the first practical applications of DAIMON’s tactile technology will likely be in hotels and convenience stores in China, where labor shortages and repetitive tasks make automation attractive. For instance, in hotels, robots could make beds, fold towels, and deliver items — tasks that require precise manipulation of soft, deformable objects. In convenience stores, robots could stock shelves, handle packaged goods, and even assist with checkout, gripping items of various shapes and fragility. These environments are semi-structured but unpredictable, making them ideal proving grounds for tactile-enabled robots. Additional early deployments may include food service, where robots need to handle foods without crushing them, and light assembly in electronics manufacturing, where fine force control is critical. The open dataset allows companies to train robots for their specific use cases, from sorting recyclables to aiding in elderly care. As the technology matures, Prof. Wang expects touch-enabled robots to enter homes, performing chores that today require human dexterity. <h2 id="question7">How does DAIMON's tactile sensor work, and what makes it special?</h2> DAIMON Robotics has developed a monochromatic, vision-based tactile sensor that is about the size of a fingertip. Inside, it contains a tiny camera that captures images of a deformable elastomer surface as it contacts an object. The sensor achieves an ultra-high resolution of over 110,000 effective sensing units — essentially a high-density touch sensor that can detect minute changes in pressure and texture. When the sensor presses against an object, the elastomer deforms, and the camera records the pattern of deformation. Advanced algorithms then reconstruct the tactile image, providing data on contact geometry, force distribution, and even slip. What makes this sensor special is its combination of resolution, compactness, and cost-effectiveness, making it suitable for large-scale deployment on robot hands. Unlike many tactile sensors that are fragile or expensive, DAIMON’s design is robust and can be manufactured in volume. The company also leverages a distributed out-of-lab collection network that can generate millions of hours of tactile data annually, feeding the Daimon-Infinity dataset and enabling continuous improvement of the AI models.

Tags:

Giving Robot Hands a Sense of Touch: DAIMON Robotics' Tactile Revolution

Related Articles