China has launched Baihu-VTouch, a groundbreaking cross-body vision-and-touch dataset containing over 60,000 minutes of robot interaction data, marking a significant leap in embodied artificial intelligence development. Released on January 27, 2026, by the National and Local Co-built Humanoid Robot Innovation Center and its technology partner, this open-source resource addresses a critical gap in tactile perception training for robots.
Breaking the Visual Dependency
While most AI systems currently rely on visual inputs, Baihu-VTouch integrates pressure and deformation measurements across 380+ task types – from delicate object handling to industrial assembly scenarios. The dataset covers 500+ real-world objects and 100+ manipulation skills, enabling robots to operate effectively in low-light environments and handle fragile items with human-like precision.
Industrial and Domestic Applications
Designed to support 90% of daily and industrial tasks, the dataset features data from wheeled and bipedal robot platforms. This aligns with China's broader robotics strategy, exemplified by the Hubei Humanoid Robot Center launched last year, which collects 10 million+ data points annually through 23 simulated environments.
Approximately 6,000 minutes of Baihu-VTouch are now accessible via OpenLoong, a robotics development platform, with researchers anticipating widespread applications in manufacturing automation and service industries.
