30,000 Hours of Tactile Data Completes Embodied AI's "Sense of Touch"! XinZhi Embodied & Fudan Release Three Reports in Succession

XinZhi Embodied, in collaboration with Fudan University, has launched its latest research results, centered on building a large-scale tactile dataset. Accumulating 30,000 hours of data, the project aims to solve the problem of robots lacking a "sense of touch" during physical interactions. Additionally, related models and data are open to the community.

Tactile perception is a key step for embodied AI to move from "seeing" to "touching". Existing visual models are relatively mature but lack fine-grained understanding of object material, softness, and friction. Introducing high-quality tactile data helps AI models better understand physical world interaction feedback and improve operation precision.

The project's decision to be fully open-source lowers the technical threshold for embodied AI R&D. Developers can directly utilize this data for model training or fine-tuning, accelerating algorithm iteration. This open strategy helps promote industry standardization and foster collaborative innovation between industry, academia, and research.

Embodied AI is currently in a transition period from the laboratory to real-world application scenarios. Data scarcity has long been one of the bottlenecks restricting its development. By accumulating long-term real interaction data, it helps train more robust general robot models, laying the foundation for future implementation in scenarios such as home services and industrial manufacturing.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.