How DAIMON Robotics Is Giving Robot Hands a Sense of Touch: An Expert Q&A

Robots today can see and hear, but they largely lack a sense of touch—a critical gap for tasks like folding laundry or assembling delicate parts. Hong Kong-based DAIMON Robotics is addressing this with their Daimon-Infinity dataset, a massive collection of tactile data aimed at making robotic hands more sensitive and dexterous. In this Q&A, we explore the technology, the team behind it, and how this initiative could transform industries from manufacturing to hospitality.

What is the Daimon-Infinity dataset and why is it significant?

Daimon-Infinity, released in April 2025, is the largest omni-modal robotic dataset for physical AI ever created. It features ultra-high-resolution tactile sensing data spanning over 80 real-world scenarios—from home chores like folding laundry to factory assembly lines. The dataset includes millions of hours of multimodal data (vision, tactile, language, and action) and captures more than 2,000 human skills. Its significance lies in its scale and diversity: by open-sourcing 10,000 hours of data, DAIMON aims to accelerate research in dexterous manipulation. Unlike previous datasets that rely mainly on vision, Daimon-Infinity elevates touch to a primary modality, enabling robots to perceive texture, pressure, and slip—critical for tasks requiring fine motor control. This could finally bridge the gap between lab demonstrations and real-world robotic deployment.

How does DAIMON's tactile sensor technology work?

DAIMON’s core hardware is a monochromatic, vision-based tactile sensor about the size of a fingertip. It packs over 110,000 effective sensing units (taxels) into a compact module. The sensor uses a camera to track deformations of a soft elastomer surface when the robot touches an object. This design provides high-resolution spatial and force information, mimicking human skin’s ability to detect contact, texture, and forces. According to the company, a distributed out-of-lab collection network can generate millions of hours of data annually, making it scalable for training AI models. This sensor is a key enabler for the Daimon-Infinity dataset, as it captures rich tactile signals that complement visual data, allowing robots to understand objects through touch—something previous datasets lack.

What is the VTLA architecture and how does it differ from VLA?

Vision-Tactile-Language-Action (VTLA) is an architecture pioneered by Prof. Michael Yu Wang and his team at DAIMON Robotics. It adds tactile sensing as a fourth modality to the common Vision-Language-Action (VLA) model. In VLA, robots primarily rely on vision to interpret scenes and language commands to plan actions, but they lack direct feedback from touch. VTLA elevates tactile information to the same level as vision, enabling robots to feel pressure, texture, and slippage during manipulation. This is crucial for tasks like gripping fragile objects or tying knots, where vision alone is insufficient. By integrating touch into the decision loop, VTLA aims to make robot manipulation more robust and adaptive in unstructured environments, such as homes or hotels.

Who is Prof. Michael Yu Wang and what is his background?

Prof. Michael Yu Wang is the co-founder and chief scientist of DAIMON Robotics. He earned his PhD at Carnegie Mellon University, studying manipulation under Matt Mason, a pioneer in robotic dexterity. He went on to found 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, Wang has spent nearly four decades advancing robotic manipulation. His vision for DAIMON Robotics stems from a critical observation: current robot manipulation lacks sensitivity due to over-reliance on vision. By championing the VTLA architecture and the Daimon-Infinity dataset, he aims to give robots a true sense of touch, transforming how they interact with the physical world.

How will the dataset accelerate real-world deployment of embodied AI?

The Daimon-Infinity dataset serves as a large-scale training resource for embodied AI models. By open-sourcing 10,000 hours of data, DAIMON provides researchers worldwide with high-quality tactile, visual, and action annotations. This removes a major bottleneck: collecting real-world manipulation data is time-consuming and expensive. With this dataset, teams can train models to generalize across diverse tasks—from picking up a wine glass to folding a shirt—using touch as a key feedback signal. Prof. Wang envisions that such models will enable robots to work reliably in dynamic settings like hotels (making beds) or convenience stores (stocking shelves) in China. The dataset also standardizes tactile benchmarks, allowing for fair comparison of different approaches, thus speeding up innovation.

Where do you see touch-enabled robots making their first real-world inroads?

Prof. Wang sees initial deployments in controlled but realistic environments such as hotels, convenience stores, and light manufacturing in China. In hotels, robots with tactile feedback can make beds, handle linens, or deliver items without dropping them. In convenience stores, they could restock shelves by feeling the texture and weight of products. The key advantage of touch is that it allows robots to adapt to variations—different materials, shapes, or positions—without prior programming. Over time, the technology could expand to healthcare (assisting in surgery) and home robotics (elderly care). The Daimon-Infinity dataset’s coverage of 80+ scenarios ensures these robots are trained for real-world clutter rather than sterile lab settings.

Why did DAIMON decide to open-source part of the data?

DAIMON’s decision to open-source 10,000 hours of data from Daimon-Infinity is strategic. While the company continues its own product development (tactile sensors and robotics), releasing this dataset fosters a global research community around tactile manipulation. It lowers the barrier for startups and universities to experiment with high-fidelity touch data, potentially leading to breakthroughs that benefit the entire field. Additionally, by setting a standard for tactile datasets, DAIMON positions itself as a leader in this niche. Open-source also attracts collaboration—partners like Google DeepMind, Northwestern University, and the National University of Singapore contributed to the project. Ultimately, more researchers using tactile data means faster adoption of touch-enabled robots, which aligns with DAIMON’s long-term goal of putting sensitive robot hands in everyday environments.