Today, as robots are fully advancing toward "embodied intelligence," how to give the steel bodies of robots a sense of touch as sensitive as humans has become the key to breaking through the bottleneck of fine operations. On January 26, 2026, the Guodi Center jointly with Weitai Robotics officially released the world's first and currently largest cross-body vision-based tactile sensor multimodal dataset - Baigu-VTouch. The emergence of this dataset not only fills the gap in the field of robot vision-tactile interaction but also provides global robot developers with a high-value digital mine.
The world's largest scale: 60,000 minutes of "real touch"
Different from previous single and fragmented experimental data, Baigu-VTouch has achieved a leap in scale and completeness:
Massive data volume: The dataset exceeds 60,000 minutes, containing approximately 90.72 million pairs of real object contact samples.
Comprehensive coverage: Integrates tactile sensor data, RGB-D depth data, and joint pose data, achieving deep alignment of multimodal information.
High-precision recording: Its sensors support a resolution of 640×480 and a high refresh rate of 120Hz, capable of capturing extremely subtle physical changes during contact.
Across-body paradigm: From "Qinglong" to "handheld terminals"
The most core breakthrough of this dataset lies in its **"cross-body"** collection characteristics. It breaks the data island of a single device, covering multiple mainstream robot configurations:
Humanoid robots: Complex dynamic structures represented by full-size humanoid robot "Qinglong."
Humanoid wheeled: Including flexible wheel-arm composite robots like D-Wheel.
Intelligent terminals: Even including handheld pika grippers and other intelligent terminal devices.
This cross-body coverage means that the algorithms trained by developers will have stronger universality, allowing robots of different forms to quickly "acquire" similar perceptual abilities.
"Matrix-style" tasks: Let robots enter everyday life
To enable AI to truly solve real-world problems, Baigu-VTouch has built an extremely systematic task matrix around four real application scenarios:
Home and domestic services: Dealing with complex interactions in home environments.
Restaurant services: Precisely handling fragile and flexible objects for grasping.
Industrial manufacturing: Achieving force control adjustments in precise assembly processes.
Special operations: Performing precise operations in extreme or special environments.
This task structure covers over 100 atomic skills and more than 260 contact-intensive tasks. Research data shows that after introducing visual-tactile perception, 68% of tasks have obtained more complete and continuous contact state descriptions, which is decisive for robots to achieve failure recovery and precise force control.
