Hesai Launches Kosmo to Turn Real-World Scans Into Robot Training Data
Hesai's new 3D data engine accelerates Physical AI development by converting physical environments into high-fidelity digital assets.
1Spatial Data Engine Launch
Hesai Group introduced Kosmo on October 14, 2026, a comprehensive platform designed to convert physical environments into high-fidelity 3D digital spaces. The system integrates the Kosmo One spatial camera with the Kosmo World cloud platform. This infrastructure supports robot training, world model development, and immersive content creation for diverse industrial applications.
2Hardware and Capture Capabilities
The Kosmo One is an 8K spatial camera featuring high-resolution lidar, a one-inch imaging sensor, and a high-precision inertial measurement unit. It records geometry, color, and texture with microsecond-level synchronization. While the maximum capture resolution is 8K, the system allows users to export data at a maximum resolution of 6K.
3Efficiency in 3D Reconstruction
Utilizing 3D Gaussian Splatting, the camera achieves millimeter-level geometric accuracy for materials and lighting. A single operator can scan a 100-square-meter indoor space in approximately 15 minutes. Hesai claims the Kosmo World platform processes this data into 3D models at speeds up to five times faster than comparable market alternatives.
4Robotics Deployment Success Metrics
In a Dairy Queen store deployment, manual modeling previously required 300 hours, which Kosmo reduced to 15 minutes of scanning. Following this transition, the Sharpa robot’s task success rate reached 99%. The robot’s failure rate dropped to one-tenth of its previous level, enabling immediate operation within the actual retail environment.
5Scaling Training Data Production
Kosmo provides precise camera poses and depth information to generate over 10,000 training trajectories from a single captured space. This capability offers developers 100 times more high-quality training data for the same cost compared to traditional methods. This efficiency is critical for building robust world models for Physical AI systems.
6Impact on Media Production
For film and television, Kosmo reduces the effort for a 200-square-meter scene from 14 person-days of CG modeling to approximately two person-days. Scanning a 10,000-square-meter streetscape takes one week instead of two months. Total production costs can fall by up to 80% through these streamlined digital asset creation workflows.
7Ecosystem and Strategic Partnerships
Hesai is building an open spatial intelligence ecosystem with integrations for NVIDIA Isaac Sim, Unreal Engine, and Pixar’s RenderMan. Strategic partnerships include Galbot, Dexmal, and Amap, with the latter aiming to shift navigation from finding stores to locating products inside them. These collaborations span robotics, filmmaking, and advanced location-based services.
8Future Growth and Actuation
Hesai views 3D world data and robotic motion as its second growth driver. The company’s actuation modules, which power the Sharpa Wave hand, reached 10,000 cumulative deliveries by mid-2026. With shipments expected to hit six figures in 2027, Hesai is expanding from hand modules to joints across the entire robot body.
Original article: Unite.AI — Hesai Launches Kosmo to Turn Real-World Scans Into Robot Training Data