DigiTimes 综述禾赛 Kosmo 三维世界数据引擎与执行器布局，但核心发布系 10 月 14 日旧闻，两地股价已连续两日下跌消化

As artificial intelligence increasingly blurs the boundary between the digital and physical worlds, complications associated with real-world deployment — particularly the extreme variability of physical environments — are emerging as a subtler configuration challenge in the industry spotlight.

While the aggressive robotics push toward more advanced forms of embodied intelligence has generated growing excitement across the industry, the promise of physical AI is increasingly running into a major technical obstacle confronting humanoid robotics: spatial awareness and understanding.
Humanoid robots may gain stronger model capabilities, but their usefulness ultimately depends on how accurately they perceive and respond to changing physical environments. That puts high-quality spatial data at the center of the next phase of physical-AI development.
Chinese lidar maker Hesai Group is targeting that bottleneck with Kosmo, its newly launched 3D world data engine, as the company attempts to expand beyond conventional lidar supply into a broader physical-AI infrastructure role.
Kosmo targets the missing data layer in physical AI
Kosmo combines an AI spatial camera with a cloud platform that converts real environments into high-fidelity 3D assets for robot training, world-model development, and other spatial-AI applications. Hesai describes the platform as infrastructure for collecting and reusing physically grounded data at scale.
Reuters reported that CEO David Li framed the expansion as an effort to position Hesai as a Tier-1 technology supplier for robotics, comparable to Bosch's role in the automotive industry.
The development of an AI spatial-intelligence platform also points to Hesai's broader ambitions beyond pure lidar supply. With actuators added to its portfolio, the company is effectively positioning itself across both the "understand" and "act" sides of physical AI.
Kosmo is designed to help close the sim-to-real gap, allowing developers to reconstruct geometrically accurate environments and generate training data without repeatedly collecting physical-world trajectories. Hesai says a single captured environment can generate more than 10,000 training trajectories, delivering over 100 times greater data-acquisition efficiency at equivalent cost under its testing assumptions.
Commercial traction is also beginning to emerge. Hesai delivered initial Kosmo prototypes in July and had secured orders from humanoid-robotics companies including Galbot, while engaging with more than 200 prospective partners across industries. Initial revenue was expected from the third quarter of 2026.
Actuators extend Hesai from perception into motion
Hesai is simultaneously expanding into robotic actuation, another high-value bottleneck in humanoid systems.
Li said actuators — which convert electrical energy into mechanical motion — can account for more than half of a robot's material costs, while demand is shifting toward smaller, lighter, and more precise components.
Hesai says its proprietary actuation modules provide roughly three times the power density of comparable products in a package 37% smaller, with transmission efficiency above 95%. Cumulative shipments exceeded 10,000 units by the end of Q2 2026, with six-figure annual shipments targeted for 2027.
Those modules are already used in Sharpa Robotics' Wave dexterous hand. In a Shanghai Dairy Queen deployment, a Sharpa robot autonomously prepared ice cream, with Hesai reporting a 99% task-success rate after Kosmo-based mapping was introduced.
That deployment provides an early demonstration of the closed-loop model Hesai is pursuing: spatial sensing creates an understanding of the environment, while actuators translate AI decisions into physical movement.
Automotive scale becomes Hesai's robotics advantage
Hesai enters the robotics market with manufacturing experience built through automotive lidar.
The company shipped 118,282 robotics lidar units in Q1 2026, up 137.8% YoY, alongside 353,441 advanced driver assistance system (ADAS) lidar units. Its first-quarter revenue reached CNY680.6 million (US$98.7 million), up 29.6% from a year earlier.
Reuters also notes that Hesai has driven automotive lidar pricing below US$200 per unit, with customers including Mercedes-Benz, Volkswagen, GM, and Ford.
The strategic question is therefore no longer whether Hesai can sell lidar beyond cars. It is whether the company can leverage automotive-scale sensing, spatial-data infrastructure, and actuation into a broader physical-AI platform — supplying not only the systems that let robots see, but increasingly the technologies that allow them to understand and act.