Runner-up · Robot training · Spatial data
Hesai maps robot training spaces in 15 minutes.
Hesai launched Kosmo, combining an 8K spatial camera, lidar, an inertial sensor and cloud reconstruction to convert real environments into reusable 3D scenes for robot training and world-model development.
Kosmo One captures geometry, color, texture and camera position with synchronized lidar, imaging and inertial data. Hesai says one operator can scan a typical 100-square-meter indoor space in about 15 minutes, after which Kosmo World produces Gaussian-splat, mesh and depth outputs with semantic segmentation and robot-camera adaptation.
Hesai linked the launch to a Dairy Queen deployment with Sharpa: a 3D environment that previously required about 300 hours of manual modeling was captured and reconstructed from a 15-minute scan, and the robot reached a reported 99% task-success rate while reducing failures to one-tenth of the earlier level. Those results come from the vendor and partners, not an independent benchmark.
Read original story ↗01
What changed
Hesai previewed Kosmo earlier in 2026. The October 14 event is the commercial launch of the integrated hardware-and-cloud workflow, with Galbot and Dexmal partnerships and a disclosed operating example connecting scene capture to a robot performing a real store task.
02
Why it matters
Robots learn poorly when training data omits the geometry, materials, lighting and exceptions of the place where they must work. Building those environments manually can make sim-to-real development slow and expensive. A repeatable capture pipeline could let manufacturers, warehouses and service operators refresh digital training spaces as layouts change, then validate behaviors before touching production. The one store example is promising but narrow: industrial adoption depends on accuracy across larger sites, handling moving equipment and occlusion, secure ownership of spatial data, update cadence and whether faster environment creation measurably reduces commissioning time and failures.
03
What to watch
Watch availability and pricing, named Galbot and Dexmal deployments, independent geometry and reconstruction benchmarks, scans of large industrial sites, change-detection workflows, spatial-data governance and whether customers report lower robot commissioning time or failure rates.
Why it was a runner-up
The system addresses a real training-data bottleneck and has one operating demonstration, but its wider industrial adoption and independent performance record remain limited.
Impact: 90/100 · Confidence: 95/100
Hesai supplies detailed hardware, workflow and deployment claims, while Reuters independently observed the launch and a robot mapping demonstration at the Shanghai factory. Confidence is high in the product launch; performance figures still rely on company and partner reporting.
Read this edition’s Daily Signal →