TARS Robotics Unveils Real-World Multimodal Dataset for Embodied AI

TARS Robotics has introduced the World In Your Hands (WIYH) dataset, described as the first large-scale, real-world Vision-Language-Tactile-Action (VLTA) multimodal dataset for embodied intelligence, announced in a press release. The dataset will be made openly accessible to research institutions and industry partners in December 2025.

WIYH is built on a first-person, human-centric data collection framework that captures real-world operational workflows across sectors such as hotel laundry, supermarket assembly, and logistics. It integrates synchronized visual, tactile, and action data with high spatiotemporal alignment and includes advanced annotations generated by in-house foundation models.

The dataset aims to overcome limitations of simulation and internet-sourced data by providing authentic, richly annotated, and large-scale multimodal information. It is designed to support the development of generalist embodied foundation models capable of performing a wide range of physical tasks across industries.