DAIMON Robotics Unveils World’s Largest Robotic Dataset with High-Resolution Touch Sensing for AI Hands

DAIMON Robotics Introduces Daimon-Infinity: A Breakthrough in Robotic Touch Sensing

In April 2024, DAIMON Robotics, a Hong Kong-based company, unveiled Daimon-Infinity, the world’s largest omni-modal robotic dataset for physical AI. This dataset integrates high-resolution tactile sensing and spans a broad spectrum of tasks, from folding laundry at home to precision tasks on factory assembly lines.

The initiative is backed by a global network of collaborators, including Google DeepMind, Northwestern University, and the National University of Singapore. The release marks a strategic milestone for DAIMON Robotics, a company established just two and a half years ago, renowned for its advanced tactile sensor hardware.

DAIMON’s Tactile Sensor Technology: A Game-Changer for Robotic Hands

DAIMON’s core innovation lies in its high-resolution tactile sensing technology, particularly its monochromatic, vision-based tactile sensor. This fingertip-sized module packs over 110,000 effective sensing units, enabling robots to perceive textures, forces, and interactions with unprecedented precision.

The company’s distributed out-of-lab data collection network generates millions of hours of data annually. This vast repository of tactile sensing data is the foundation for building large-scale robot manipulation datasets. To further accelerate the adoption of embodied AI, DAIMON has made 10,000 hours of its data open-source.

Vision-Tactile-Language-Action (VTLA): Redefining Robotic Perception

At the heart of DAIMON’s strategy is Prof. Michael Yu Wang, co-founder and chief scientist of the company. Prof. Wang, a pioneer in robotics, holds a PhD from Carnegie Mellon University, where he studied manipulation under Matt Mason. He later established the Robotics Institute at the Hong Kong University of Science and Technology.

An IEEE Fellow and former Editor-in-Chief of IEEE Transactions on Automation Science and Engineering, Prof. Wang has dedicated nearly four decades to the field. His vision is to address the critical gap in robot manipulation: the lack of tactile sensitivity. Traditional models rely heavily on Vision-Language-Action (VLA), often overlooking the importance of touch. Prof. Wang and his team have pioneered the Vision-Tactile-Language-Action (VTLA) architecture, positioning tactile feedback as an equally vital modality alongside vision.

"Our objective is to address the missing ‘insensitivity’ of robot manipulation, which practically relies on the dominant Vision-Language-Action (VLA) model."

Daimon-Infinity: The World’s Largest Omni-Modal Dataset for Physical AI

Daimon-Infinity stands out as the largest and most comprehensive robotic manipulation dataset to date. It features:

• Million-hour scale multimodal data;
• Ultra-high-resolution tactile feedback;
• Data collected from 80+ real-world scenarios;
• Insights into 2,000+ human skills.

This dataset is designed to enhance our understanding of robotic hands operating in natural environments, bridging the gap between laboratory research and real-world applications.

Why DAIMON Released Its Dataset Now

DAIMON Robotics has spent the past two and a half years developing high-resolution, multimodal tactile sensing devices. These devices, particularly its fingertip-sized sensors, have gained recognition and adoption among academic institutions, research labs, and leading humanoid robotics companies.

As embodied AI continues to evolve, the role of high-quality data has become increasingly critical. By releasing Daimon-Infinity, DAIMON aims to:

• Accelerate the development of embodied intelligence;
• Enable researchers and developers to train more dexterous and adaptive robots;
• Foster collaboration across the global robotics community.

Prof. Wang emphasizes that the dataset’s release is a strategic move to democratize access to high-fidelity tactile data, which has historically been scarce. This initiative is expected to unlock new possibilities in robotics, from household assistance to industrial automation.

Real-World Applications: Where Touch-Enabled Robots Will First Appear

When asked about the first real-world applications of touch-enabled robots, Prof. Wang highlights sectors in China where these technologies are poised to make an impact:

• Hotels: Robots equipped with tactile sensing could assist in tasks like bed-making, room cleaning, and handling delicate fabrics;
• Convenience stores: Automated systems with tactile feedback could manage inventory, restock shelves, and handle cashier duties with precision;
• Manufacturing and logistics: Robots could perform intricate assembly tasks, quality control, and packaging with enhanced dexterity.

These applications demonstrate the transformative potential of tactile sensing in making robots more adaptable to dynamic and unstructured environments.