Hesai Launches Kosmo to Accelerate Physical AI Training

The system centers on the Kosmo One camera, which utilizes lidar, a one-inch imaging sensor, and an inertial measurement unit to capture high-fidelity geometric and texture data. Using 3D Gaussian Splatting, the platform transforms physical locations into digital assets at speeds reportedly five times faster than existing market alternatives. For developers, this provides a scalable way to generate simulation-ready environments for embodied AI, shifting the workload from manual 3D modeling to automated capture and processing.

Hesai is positioning the tool as a foundational layer for both robotics and creative industries. Real-world applications have already begun, including a deployment at a Dairy Queen store where the platform reduced 300 hours of manual modeling to a 15-minute scan, ultimately improving a robot’s task success rate to 99%. Beyond robotics, the technology is being utilized by film studios to streamline virtual production and by cultural heritage organizations to digitize historical sites. Through strategic partnerships with firms like Amap and Pixar’s RenderMan, Hesai intends to integrate these spatial data workflows into broader commercial and media ecosystems, supporting its transition from a pure lidar manufacturer to a comprehensive Physical AI infrastructure provider.