Hesai launches Kosmo 3D world data engine with AI spatial camera for robot training
Hesai has launched Kosmo, a 3D world data engine that pairs an 8K AI spatial camera with a cloud platform to turn real places into 3D training data for robots and world models.
- ~15 minutes
- ~300 hours
- 99%
- up to 80%
What Happened
Hesai Group launched Kosmo, a 3D world data engine it says is built to solve one of the most critical bottlenecks in Physical AI: access to high-quality real-world data at scale. Kosmo combines Kosmo One, described as the world's first AI spatial camera, with the Kosmo World cloud platform to transform physical environments into high-fidelity 3D digital spaces for robot training, world model development and immersive content creation. At the launch event, Hesai also announced strategic partnerships with embodied AI companies Galbot and Dexmal and showcased applications developed with Pixar's RenderMan and Amap, demonstrating Kosmo's potential across robotics, filmmaking and location services.
Kosmo One, which Hesai calls the world's first 8K ultra-high-definition AI spatial camera, integrates high-resolution lidar, a one-inch imaging sensor and a high-precision inertial measurement unit to capture geometry, color, texture and camera position with microsecond-level synchronization. Its 3D Gaussian Splatting reconstruction delivers millimeter-level geometric accuracy and reproduces materials, lighting and shadows. The camera offers instant scanning on power-up, intelligent guidance and one-click scan resumption, so a single operator can scan a typical 100-square-meter indoor space in about 15 minutes. Kosmo World, powered by Hesai's Kosmo Spatial Core, generates 3DGS, mesh and depth data in a single reconstruction, uses generative AI to fill unscanned areas and rebuild materials and textures, and applies AI-powered semantic segmentation to identify individual objects; views can be adapted to a robot's camera characteristics to narrow the simulation-to-reality gap. Hesai is building Kosmo as an open ecosystem, with open access to raw capture data and 3D assets and integration with NVIDIA Isaac Sim, Isaac Lab, Unreal Engine, RenderMan and Houdini, under two business models: integrated hardware and software for robotics, media and other industries, and services that produce high-quality real-world spatial data at scale for world model and Physical AI developers.
- ~300 hours of manual modeling by a professional team
- 15 minutes
- 99% task success rate; failure rate down to one-tenth of the previous level
- Film and media: a location can be captured once and reused from previsualization and virtual production to postproduction, with new camera angles, lighting changes and reusable scene elements in tools such as Unreal Engine. A 10,000-square-meter European streetscape that takes about two months to build as a physical set can be scanned, reconstructed and edited in about one week, with potential cost savings of up to 80%; for a 200-square-meter indoor scene, traditional CG modeling takes about 14 person-days versus about two with Kosmo.
- Film partners include leading film and television companies using workflows built around Pixar's production-grade renderer RenderMan, which has released an official tutorial and downloadable resources; Hesai is also partnering with Yuemo Studio on AI-assisted filmmaking that pairs captured real-world environments with AI-generated content to reduce token consumption.
- Cultural tourism and heritage: Hesai is working with TRIC, a United Nations partner, to create an immersive digital artwork of Du Fu Thatched Cottage in Chengdu, the former home of Tang Dynasty poet Du Fu.
- Spatial commerce: through a partnership with Amap, reconstructions of the Gravity Art Museum and PIFO Gallery are already live in Amap's Flying Street View feature; Amap reaches nearly one billion monthly active users, and the aim is to extend navigation from finding a store to locating a product inside it.
Hesai is extending more than a decade of expertise in lidar, chip development and precision manufacturing into the broader Physical AI market, building its next phase of growth around 3D world data and robotic motion. Its lidar deliveries have doubled for nine consecutive years, and it holds a leading global market position in passenger-vehicle ADAS and robotaxis. Alongside Kosmo, Hesai is expanding its robotic actuation business: its actuation modules power Sharpa's Wave dexterous hand and deliver three times the power density of comparable products in a package 37% smaller. The hand has been incorporated into NVIDIA's Isaac GR00T humanoid robot reference design and adopted by research institutions including UC Berkeley and Stanford; cumulative deliveries of Hesai's actuation modules exceeded 10,000 units by the end of the second quarter of 2026, and shipments are expected to reach six figures in 2027.
“Physical AI needs real-world data that is accurate, affordable and available at scale. Meeting all three requirements has been a persistent problem. Kosmo is designed to overcome that challenge, making it easier for AI developers and creators to turn real environments into data they can train on and assets they can build with.”
“Physical AI needs infrastructure that lets it understand the real world and act within it. Our 3D world data engine and robotic motion engine address both needs, building directly on Hesai's strengths to create our second growth driver. Over the past decade, we have enabled machines to see the world. Over the next decade, we will enable robots to understand and change it.”
Previously from Hesai Technology
Hesai has recently been pushing into high-precision 3D mapping: in September 2026 it launched its MT series lidar at INTERGEO 2026 in Germany, promising ultra-high spatial resolution and 3.5 mm ranging precision in what it calls the smallest form factor in its class. The company has also published a plain-language guide to how lidar works, explaining how laser pulses become 3D point clouds, why the technology still functions in darkness, and where it goes next.
- Hesai launches MT series lidar for high-precision 3D mapping, claiming smallest form factor in its class
- Hesai explains how lidar 3D perception works, from point clouds to full-color sensing
Background drawn from MotorClaw's earlier coverage of Hesai Technology's official releases.
Why this matters
Robot and world-model developers have long struggled to get accurate, affordable real-world 3D data at scale, and Kosmo is aimed squarely at that bottleneck. If it performs as described, it could sharply cut the time and cost of building robot training environments — Hesai says a job that took about 300 hours of manual modeling was finished in 15 minutes. It is also Hesai's bid for a second growth driver beyond selling lidar to automakers.
Terms in This Story
- 3D Gaussian Splatting (3DGS)
- A 3D reconstruction technique that represents a scanned scene as many small, overlapping blobs, allowing detailed, lifelike models to be generated from captured sensor or camera data.
- lidar
- A sensor that measures distance by firing laser pulses and timing how long they take to return, building a 3D picture of the surrounding environment.
- embodied AI
- Artificial intelligence that controls a physical body, such as a robot, so it must perceive and act in the real world rather than only in software.
- world model
- An AI system that builds an internal representation of an environment so it can predict how that environment behaves and plan actions within it.
Summarised from the linked release; details can be imperfect — always verify against the original source.