Hesai, a lidar company known for automotive sensors, has launched Kosmo, a 3D world data engine aimed at one of Physical AI's persistent bottlenecks: high-quality real-world data at scale. The stack pairs Kosmo One, which Hesai calls the world's first 8K ultra HD AI spatial camera, with the Kosmo World cloud platform to turn physical spaces into high-fidelity 3D digital environments for robot training, world-model development and immersive content.
Kosmo One combines high-resolution lidar, a one-inch imaging sensor and a high-precision inertial measurement unit (IMU). Hesai says the sensors stay synchronized at the microsecond level so the device captures geometry, color, texture and camera pose together. Reconstruction uses 3D Gaussian Splatting (3DGS), which the company says can reach millimeter-level geometric accuracy and reproduce materials, lighting and shadows. A single operator can scan a typical 100-square-meter indoor space in about 15 minutes, according to Hesai, and Kosmo World then builds 3D models at speeds the company puts at up to roughly five times those of comparable products. Hesai notes that "8K" refers to maximum capture resolution, while exported files top out at 6K.
Kosmo World is powered by Hesai's Kosmo Spatial Core (KSC). In one reconstruction pass it produces 3DGS, mesh and depth data in a single consistent scene. Generative tools can fill unscanned areas and rebuild materials, while semantic segmentation labels individual objects. Views can also be adapted to a robot's camera characteristics to narrow the gap between simulation and the real world.
A Dairy Queen ice cream store deployment illustrates the pitch. Hesai says building a 3D training environment there previously took about 300 hours of manual modeling; with Kosmo, scanning and reconstruction took 15 minutes. In that setup, the company says a Sharpa robot reached a 99% task success rate and cut its failure rate to one-tenth of the previous level. Separately, Hesai says a single captured space can automatically yield more than 10,000 training trajectories, delivering more than 100 times as much high-quality training data for the same cost. Those figures are Hesai's own claims.
Beyond robotics, Kosmo plugs into an open ecosystem that includes NVIDIA Isaac Sim and Isaac Lab, Unreal Engine, Pixar's RenderMan and Houdini. Hesai announced partnerships with embodied AI firms Galbot and Dexmal, is working with Amap on Flying Street View reconstructions already live for venues such as the Gravity Art Museum and PIFO Gallery, and is collaborating with Yuemo Studio on AI-assisted filmmaking and with TRIC on a digital preservation project for Du Fu Thatched Cottage in Chengdu.
SEE ALSO: Hesai's Three Co-Founders Launch New General Robotics Startup Sharpa · Sharpa Unveils D01 Tactile Robot, W02 Dexterous Hand and AE01 Haptic Glove at IROS 2026 · Galbot Prices ET1 Bipedal Humanoid From 79,000 Yuan, With Watch-and-Learn Skills on Upper Tiers