RoboSense, GenRobot.AI team up on real-world data for physical AI

Gasgoo Munich- RoboSense and GenRobot.AI announced a strategic partnership on July 19 during the 2026 World Artificial Intelligence Conference in Shanghai, seeking to address one of embodied AI's central challenges: acquiring sufficiently rich and accurate real-world training data.

The collaboration will pair RoboSense's 3D sensing technology with GenRobot.AI's data infrastructure for embodied intelligence. Initial work will focus on collecting multi-view egocentric data and building datasets that combine several modalities with three-dimensional representations of physical environments. The companies said the resulting infrastructure would support the development of physical AI systems, which must learn to perceive and interact with the real world.

Image source: RoboSense

GenRobot.AI specializes in first-person human-action data for training embodied AI models. Its technology stack covers data capture, hand-pose reconstruction, 3D ground-truth annotation and standardized delivery of multimodal datasets.

At the center of the system is the company's proprietary Data Foundation Model, or DFM, which operates as part of a closed-loop data pipeline. GenRobot.AI says the platform can maintain hand-tracking accuracy of under one centimeter and reconstruct hand poses with millimeter-level precision.

Its Gen DAS hardware captures synchronized egocentric data from multiple viewpoints, covering movements of the head, hands and full body. The company has also established an automated data-production system known as Gen ADP. According to GenRobot.AI, the system has been deployed across more than 10,000 real-world settings in areas including homes, manufacturing, logistics and healthcare, generating a cumulative data pool of more than one million hours of human activity.

RoboSense, best known for its lidar and perception technologies, has been expanding from automotive sensing into a broader robotics platform spanning sensors, proprietary chips and embodied AI solutions. Its products are used in more than a dozen robotics categories, including humanoid and quadruped robots as well as autonomous commercial cleaning machines. The company describes itself as the world's largest supplier of lidar products for robotics, a position it attributes to third-party shipment data.

RoboSense's sensing hardware captures information such as distance, depth, position, orientation and spatial structure. Such inputs allow robots to construct representations of their surroundings, learn how objects and environments behave, and plan more complex physical tasks. The company's current portfolio and physical AI strategy also extend from environmental sensing and data acquisition to decision-making and manipulation.

Under the partnership, the two companies plan to connect sensing hardware more closely with the full data lifecycle, from collection and processing to model training and deployment. The goal is to give robotic systems more reliable data for understanding their surroundings, acquiring new skills and adapting to changing conditions.

The agreement reflects a wider shift in embodied AI development. As robotics companies move beyond model design toward deployment in unstructured environments, the availability and quality of real-world interaction data are becoming as important as computing power and algorithms. RoboSense and GenRobot.AI are betting that a more integrated data pipeline can help shorten that path from training to commercial use.