DAIMON Robotics has made a surprising move by deciding to open-source 10,000 hours of tactile data from its Daimon-Infinity project. This is an effort to break down the biggest barrier in physical AI: the ability to delicately sense and interact with real-world objects.
Background
While current AI models primarily rely on vision and language (VLA), DAIMON proposes the VTLA (Vision-Tactile-Language-Action) model. Their core technology is a high-resolution optical tactile sensor that can pack 110,000 sensing units into a single robotic fingertip. The Daimon-Infinity dataset was gathered from a distributed network across 80 real-world scenarios, in collaboration with major partners such as Google DeepMind and the National University of Singapore.
Why it matters
Having a tactile sense will help robots perform dexterous tasks such as folding clothes, holding eggs, or assembling electronic components without damaging the objects. For robotics research teams in Vietnam, DAIMON's open-sourcing of this large-scale dataset is a golden opportunity to train robot control models without the high costs of expensive physical data collection. This is an important preparatory step for the era of service robots as dexterous as human hands.