Today, with the rapid development of embodied intelligence, how to enable robots to have a sensitive "tactile" perception like humans has become the key to industry breakthroughs. On January 27, 2026, the National-Local Co-built Humanoid Robot Innovation Center, in collaboration with multiple research teams, officially released a cross-body visual-tactile multimodal dataset named "Baihu-VTouch." This achievement not only fills the gap in large-scale visual-tactile data but also provides global robot developers with a highly valuable "digital mine."
Massive Data Accumulation: Over 60,000 Minutes of "Sensory Memory"
As one of the largest open-source datasets of its kind globally, Baihu-VTouch has achieved a qualitative leap in both breadth and depth of data:
Extended Interaction Duration: It includes over 60,000 minutes of real robot interaction data.
Multi-dimensional Sensory Alignment: The dataset deeply integrates multimodal information such as visual footage, tactile feedback, and robotic joint poses.
Physical Property Capture: Using high-precision sensors, it allows AI to learn subtle physical changes and deformation logic during object contact.
Cross-Body Paradigm: Breaking the "Perception Barrier" Between Hardware
The most notable feature of this dataset is its "cross-body" characteristic. It is no longer limited to a single robot model but includes sensory data from different configurations (such as humanoid robots, wheeled robots, and robotic arms). This general-purpose data architecture enables AI models to quickly transfer and generalize perception across different hardware, allowing various types of robots to rapidly acquire fine manipulation capabilities.
Empowering Embodied Intelligence: From "Seeing" to "Understanding"
For a long time, robot operations have relied heavily on vision. However, in situations involving transparent objects, dimly lit environments, or precise assembly tasks, vision often "fails." The release of "Baihu-VTouch" marks the evolution of robots from a "vision-dominated" approach to a new stage of "visual-tactile integration." This dataset will provide a more solid foundation for scenarios such as home services, industrial precision manufacturing, and medical assistance.
With the open-source release of this dataset, the global robotics field is expected to experience a wave of perception algorithm iteration. When robot fingertips are no longer cold and unresponsive, an era of truly dexterous embodied intelligence is accelerating toward us.
