/

Hesai Kosmo

Hesai Kosmo

Social Media Links

Linkedin
Twitter
YouTube

Hesai Kosmo pairs the Kosmo One lidar and 8K spatial camera with the Kosmo World cloud, which reconstructs 3DGS, mesh and depth data for robotics and film.

Location

Shanghai, Cn

Platform

Capture hardware (Kosmo One lidar and 8K spatial camera) with the Kosmo World cloud platform for reconstruction, management and export

GPU

None published. Hesai says reconstruction runs in its Kosmo World cloud, which implies no local GPU is needed to generate splats. Rendering exported splats in an engine or DCC uses that host's own GPU

Pricing

Not published. Sold as integrated hardware and software, or as a real world spatial data production service. Buyers go through a contact sales form

License

Proprietary

Compare licence terms ↗

Best for

Lidar assisted 3DGS, mesh and depth capture of real spaces as robot training and world model data, and for film, virtual production and mapping

Updated

July 2026

Kosmo is a 3D world data engine from Hesai Technology, the Shanghai lidar maker listed on Nasdaq and the Hong Kong stock exchange. Hesai launched Kosmo on October 14, 2026 as two products, the Kosmo One capture device, which Hesai calls an AI spatial camera, and Kosmo World, a cloud platform that turns captures into 3D Gaussian Splatting scenes, meshes and depth data. Hesai positions Kosmo first as a source of real world 3D training data for robots and world models, and also as a capture tool for film, cultural tourism and mapping.

Hesai previewed Kosmo at its Technology Open Day on April 17, 2026, alongside the Picasso full color SPAD SoC and the next generation ETX automotive lidar. In July Hesai showed Kosmo at WAIC 2026 in Shanghai, billing it as the world's first AI spatial camera and setting its launch for the second half of 2026. Hesai reported in its second quarter 2026 results that it delivered initial prototypes in July 2026 and secured orders from humanoid robotics companies including Galbot.

Kosmo One combines a lidar, an 8K imaging system with a one inch sensor and a high precision IMU, which Hesai says are synchronized to the microsecond so geometry, color, texture and device pose are recorded together. Hesai calls it the world's first AI spatial camera with an integrated 8K imaging system, a claim it dates to October 2026, and says its 3DGS reconstruction reaches millimeter level geometric accuracy. 8K is the maximum capture resolution, and exported files top out at 6K. The Kosmo product page lists a Class 1 lidar at 1535 nm with a 320 by 60 degree field of view, 569,000 points per second and 40 meters of range at 10 percent reflectivity, along with a weight of about 1,200 grams including the battery handle, 512 GB of internal storage and about 60 or 120 minutes of runtime from the standard and extended batteries. Hesai says one operator can scan a typical 100 square meter indoor room in about 15 minutes, with actual time depending on the space.

Reconstruction runs in Kosmo World on what Hesai calls Kosmo Spatial Core, which produces aligned 3DGS, mesh and depth outputs from a single reconstruction. For robot training Hesai assigns the splats to visual realism, the mesh to interactive objects and the depth to spatial distance. Generative AI fills areas the scan missed and reconstructs materials and textures, semantic segmentation labels individual objects so a scene can serve as sim ready data, and rendered views can be matched to a target robot's camera. Hesai says Kosmo World generates models up to about five times faster than comparable 3DGS products under similar hardware, data and quality conditions. In one deployment Hesai describes, a Dairy Queen store that would have taken about 300 hours to model by hand was scanned and reconstructed in about 15 minutes as a training environment for a Sharpa robot, which reached a 99 percent task success rate.

Hesai offers open access to raw capture data, including images, point clouds and IMU data, alongside 3DGS and mesh assets, and the product page lists .sog and .ply for splats, .obj for meshes, and .usd and .usdz. Hesai names NVIDIA Isaac Sim, Isaac Lab, Unreal Engine, RenderMan and Houdini as integrations. Pixar published the RenderMan tutorial Gaussian Splatting the Pixar Campus on October 5, 2026, with a downloadable Houdini Solaris project built on a Kosmo scan of the Pixar campus and a script that converts uncompressed splat PLY files into a native splat primitive, tested with Houdini 22.0.429 and the RenderMan 28 beta ahead of splat support shipping in RenderMan 28. Pixar's Dylan Sisson describes the same pipeline in A Brush with Reality, capturing the Pixar plaza and the studio's Lab with Kosmo and bringing the plaza into Solaris through USD, then reshaping the splats with standard Houdini point tools and a camera based falloff before rendering them in RenderMan.

At launch Hesai announced strategic partnerships with the embodied AI companies Galbot and Dexmal, and showed work with Amap, whose Flying Street View feature already carries Kosmo reconstructions of the Gravity Art Museum and PIFO Gallery. Hesai sells Kosmo two ways, as integrated hardware and software solutions and as a service producing real world spatial data at scale for world model and physical AI developers. Hesai has not published a price or a general shipping date, and the Kosmo site routes buyers to a contact sales form. Other capture hardware makers whose software turns lidar and imagery into Gaussian splats include XGRIDS, CHCNAV and Artec 3D.

Frequently Asked Questions

What is Hesai Kosmo?

Hesai Kosmo is a 3D world data engine from the lidar maker Hesai Technology, launched on October 14, 2026. Kosmo pairs the Kosmo One capture device, which combines lidar, an 8K imaging system and an IMU, with the Kosmo World cloud platform, which reconstructs captured spaces as 3D Gaussian Splatting scenes, meshes and depth data.

Does Hesai Kosmo create Gaussian splats?

Yes. Hesai Kosmo reconstructs scans with 3D Gaussian Splatting (3DGS) in the Kosmo World cloud, where Hesai's Kosmo Spatial Core produces aligned 3DGS, mesh and depth outputs from one reconstruction. Hesai says Kosmo reaches millimeter level geometric accuracy and that exported files go up to 6K, although the camera captures at 8K.

When is Hesai Kosmo available?

Hesai officially launched Kosmo and the Kosmo One camera on October 14, 2026, after previewing it in April 2026 and delivering prototypes in July 2026. Hesai has not published a general shipping date or regional availability, and the Kosmo website routes buyers to a contact sales form.

How much does Hesai Kosmo cost?

Hesai has not published a price for Kosmo or Kosmo One. Hesai sells Kosmo as integrated hardware and software solutions and, separately, as a service that produces real world spatial data at scale for world model and physical AI developers.

What is Hesai Kosmo used for?

Hesai Kosmo is aimed first at producing real world 3D training data for robots, embodied AI and world models, and Hesai has announced strategic partnerships with Galbot and Dexmal. Hesai also shows Kosmo in film and virtual production, in cultural tourism, and in mapping through Amap's Flying Street View.

What formats does Hesai Kosmo export, and which tools does it work with?

Hesai Kosmo exports raw capture data, including images, point clouds and IMU data, along with 3DGS and mesh assets, and the Kosmo product page lists .sog and .ply splats, .obj meshes and .usd and .usdz files. Hesai names NVIDIA Isaac Sim, Isaac Lab, Unreal Engine, RenderMan and Houdini as integrations, and Pixar's RenderMan team has published a tutorial that renders a Kosmo scan of the Pixar campus in RenderMan through Houdini Solaris.

Alternatives