Hesai Group published a press release on OCTOBER 14, 2026, announcing KOSMO, a 3D world data engine that pairs an AI spatial camera called Kosmo One with a cloud platform called Kosmo World. The company frames the product as INFRASTRUCTURE for Physical AI: high-fidelity digital twins of real rooms that ROBOTS can TRAIN on, world models can LEARN from, and film or location teams can REUSE as immersive ASSETS.
The announcement sits on Hesai’s own site and on PR Newswire, with a consumer-facing product page at kosmoai.com. Metaverse Watcher had NO prior Kosmo coverage. The company CLAIMS Kosmo One is the world’s FIRST AI spatial camera with an INTEGRATED 8K imaging system, based on publicly available product information as of October 2026. Exported files, Hesai notes, TOP OUT at 6K.
What Kosmo One Captures
Kosmo One COMBINES high-resolution lidar, a one-inch imaging sensor, and a high-precision IMU. Hesai says the sensor stack records geometry, color, texture, and camera POSITION with microsecond-level sync. Reconstruction uses 3D Gaussian Splatting (3DGS) and is pitched as MILLIMETER-level geometric ACCURACY that preserves MATERIALS, LIGHTING, and SHADOWS.
On the practical side, Hesai says a SINGLE OPERATOR can SCAN a typical 100-square-meter indoor space in about 15 MINUTES, with instant scanning on power-up, guidance during capture, and one-click scan RESUMPTION. Kosmo World then processes the capture into models that Hesai rates as up to roughly FIVE TIMES FASTER than comparable 3DGS products under similar hardware and quality settings. Those SPEED figures are COMPANY estimates with the usual caveats about scene size and parameters.
Robot Training, Not Just Pretty Scans
Kosmo World is described as a one-stop platform that emits 3DGS, MESH, and DEPTH from a SINGLE reconstruction, then fills unscanned areas and labels objects with generative and semantic tools. Views can be ADAPTED to a robot’s camera characteristics, which is Hesai’s answer to the sim-to-real gap.
The headline DEPLOYMENT is a Dairy Queen ice cream STORE. Creating a 3D TRAINING environment previously took about 300 HOURS of MANUAL modeling, Hesai says. With Kosmo, SCANNING and RECONSTRUCTION took 15 minutes. In that deployment, a Sharpa robot’s task SUCCESS RATE reached 99%, and the FAILURE RATE fell to one-tenth of the prior level, enough for the ROBOT to BEGIN operating in a REAL STORE. From a single captured space, Kosmo can also auto-generate more than 10,000 training TRAJECTORIES, which Hesai translates into more than 100 TIMES the high-quality TRAINING DATA for the SAME COST versus collecting one real-robot TRAJECTORY.
Immersive Media, Heritage, and Maps
Beyond robotics, Hesai sells Kosmo into film and media, cultural heritage, and SPATIAL commerce. Artists can pull captures into Unreal Engine for new angles and lighting. A 10,000-square-meter European streetscape that would take TWO MONTHS as a physical set can be scanned, reconstructed, and edited in about ONE WEEK, with potential cost savings of up to 80%, the company estimates. Workflows with Pixar’s RenderMan are already in the demo reel, and RenderMan has published an official tutorial. Hesai is also working with TRIC, a United Nations partner, on an immersive digital artwork of Du Fu Thatched Cottage in Chengdu, and with Amap so that reconstructions of the Gravity Art Museum and PIFO Gallery appear in Amap’s “Flying Street View.”
The stack opens RAW capture DATA and 3D assets to developers, with listed INTEGRATIONS for NVIDIA Isaac Sim, Isaac Lab, Unreal Engine, RenderMan, and Houdini. Strategic PARTNERSHIPS with Galbot and Dexmal were announced at the same event. Separately, Hesai said its actuation modules power Sharpa’s Wave hand, sit in NVIDIA’s Isaac GR00T reference design, and had PASSED 10,000 cumulative DELIVERIES by the end of Q2 2026.
What to Watch
Kosmo is a VENDOR LAUNCH, not an independent benchmark. The 8K, 15-minute, five-times, 99%, and 100× figures all come from Hesai’s notes and should be read as CLAIMS until third parties measure them. For XR and spatial computing readers, the INTERESTING part is the PIPELINE: lidar plus CAMERA capture into EDITABLE 3DGS WORLDS that already TARGET Unreal, RenderMan, and robot SIMULATORS. If the scan-to-asset loop holds up outside demo stores, Kosmo becomes another path from REAL ROOMS to IMMERSIVE content and TRAINING data, not just another lidar SKU.
Sources: Hesai launch post; PR Newswire; Kosmo product site. Related discussion: Robot Companies on X.