Hesai launches Kosmo 3D world data engine for robot training
Hesai says Kosmo can turn real spaces into reusable 3D environments for robot training, world models and immersive content.
Hesai has launched Kosmo, a 3D world data engine designed to turn real places into detailed digital environments for robot training, world models and immersive content.
The company unveiled Kosmo at its AI Day in Shanghai on Oct. 14. TechNode reports that the platform combines lidar, imaging and motion data with AI-generated scene completion and object labelling, creating reusable 3D environments from physical spaces. 1
Hesai says the problem Kosmo is targeting is the shortage of reliable real-world data for Physical AI. Digital systems can learn from vast online datasets, but robots need to understand the geometry, obstacles and constraints of places they must physically navigate. The company presents Kosmo as infrastructure for capturing and converting those places into training material. 2
From a room to a training world
According to the launch material, Kosmo can capture a space with lidar, imaging and motion data before turning it into a high-fidelity 3D scene. One capture can generate more than 10,000 training trajectories, while the resulting assets can be adapted for robot simulation and world-model development. 1
Hesai showcased a Dairy Queen store demonstration with robot maker Sharpa, where the company said a 15-minute Kosmo scan replaced about 300 hours of manual modelling. The reconstructed setting was then used to demonstrate a 99% task-success rate. Those figures are company claims from the launch coverage, not an independent benchmark. 1
The platform also reaches beyond robotics. Hesai says Kosmo assets can support visual effects, cultural tourism, heritage projects and spatial commerce, with integrations including NVIDIA Isaac Sim, Isaac Lab, Unreal Engine, RenderMan and Houdini. The company has previously described Kosmo as a spatial-intelligence platform built around an AI spatial camera, 3D assets and cloud services. 2
Our opinion
Kosmo is an interesting answer to a very real bottleneck: robots cannot learn the physical world from webpages alone. The harder question is whether captured scenes remain accurate, varied and affordable enough to matter outside polished demonstrations. A large library of digital spaces is useful only if it reflects the messiness of the real ones.
Hesai’s launch is therefore less about a single camera than about a proposed data pipeline. If Kosmo can make real environments easier to capture, edit and replay, it could give robotics developers a practical bridge between simulation and deployment. The next proof point will be repeatable results from independent teams, not just impressive scans at launch events.
- Hesai launched Kosmo as a 3D world data engine for Physical AI and robot training.