Shanghai, October 14th / PRNewswire / -- Hesai Group ( Nasdaq : HSAI ; HKEX : 2525) (referred to as “the Company” today) released Kosmo, a 3D world data engine designed to address a key bottleneck in Physical AI: the large-scale acquisition of high-quality real-world data. Kosmo consists of Kosmo One and Kosmo World cloud platforms. The former is the world's first AI spatial camera, while the latter can transform the physical environment into a high-fidelity 3D digital space for robot training, world model development, and immersive content creation. At the press conference, Hesai also announced strategic partnerships with AI companies Galbot and Dexmal, and demonstrated an application jointly developed with Pixar's RenderMan as well as with Gaode Maps, showcasing the potential of Kosmo in fields such as robotics, film and television production, and local lifestyle services.

Digitize the real world to provide data for Physical AI.

"Physical AI requires accurate, economical, and scalable real-world data," said Xiang Shunqing ( Shaoqing Xiang ), co-founder of Hesai and also CTO. "Meeting all three of these requirements has always been a long-standing challenge. The design of Kosmo is aimed at overcoming this challenge, making it easier for AI developers and creators to transform real environments into trainable data and assets that can be used for construction."

Kosmo One is the world's first 8K ultra-high-definition AI spatial camera, capable of capturing real-world environments as detailed and reusable 3D digital assets. It integrates high-resolution lidar, one-inch imaging sensors, and high-precision inertial measurement units (IMU) to synchronously capture geometry, color, texture, and camera position at the microsecond level. The advanced 3D Gaussian Splatting (3DGS) reconstruction technology achieves millimeter-level geometric accuracy and realistic details, faithfully restoring materials, lighting, and shadows.

For ease of use, Kosmo One supports instant scanning upon startup, intelligent guidance, and one-click renewal of scanning. A single operator can scan a typical 100-square-meter indoor space in about 15 minutes. Subsequently, Kosmo World can process the collected data into high-quality 3D models, at a speed that is approximately 5 times faster than that of similar products.

Turn the real world into a training ground for AI.

Kosmo World is a one-stop platform for 3D content creation and management that can convert collected data into a simulation environment suitable for AI training. This platform is driven by Hese's Kosmo Spatial Core ( KSC ), and is capable of generating 3DGS, meshes, and depth data in a single reconstruction, presenting visual realism, interactive objects, and precise spatial information within the same scene. Generative AI can fill in unscanned areas and reconstruct materials and textures, while the semantic segmentation driven by AI can identify individual objects. The perspective can also be adapted to the characteristics of robot cameras, thereby narrowing the gap between simulation and real-world operation.

In a deployment case at a Dairy Queen ( DQ ) ice cream store, this capability was verified. Previously, creating a 3D training environment required a professional team to spend about 300 hours on manual modeling; with the use of Kosmo, scanning and reconstruction took only 15 minutes. In this deployment, the task success rate of the Sharpa robots reached 99%, and the failure rate dropped to one-tenth of what it was before, allowing the robots to begin operating in a real store.

Let world models take root in reality

Kosmo also provides a real-world foundation for the development of world models. Its geometrically precise 3D scenes integrate accurate camera pose, depth information, and consistent multi-angle views, offering a real-world reference for developers to create realistic virtual environments. Based on a single collection space, Kosmo can automatically generate over 10,000 training trajectories, allowing developers to obtain more than 100 times the amount of high-quality training data at the same cost.

In addition to embodying AI and world models, Kosmo can also bring value to a growing range of fields, including film and television media, cultural tourism and heritage, as well as the commercial application of spatial data.

Film and television media: faster production, greater creative freedom

Kosmo allows film producers to collect location data just once, which can then be reused across various stages of production, from preliminary rehearsals and virtual production to final editing. In mature tools like Unreal Engine, artists can explore new shooting angles, adjust lighting, and reuse elements from a single scene. Building a 10,000-square-meter European street scene with physical sets would take two months, but with scanning, reconstruction, and editing, it can be done in just one week, potentially saving up to 80% in costs.

Global leading film and television companies are already using Kosmo, including workflows built around the movie-grade renderer RenderMan for Pixar. RenderMan allows artists to precisely control the appearance and stylization effects of the capture environment. To help creators adopt this workflow, Pixar's RenderMan has also released official tutorials and downloadable resources.

Hesai is still collaborating with Yuemo Studio to carry out AI for auxiliary film production. By combining the collected real-world environments with the content generated by AI, they aim to reduce token consumption while maintaining a high level of visual realism.

Cultural Tourism and Heritage: Preserving History in 3D

Kosmo allows cultural heritage to be permanently preserved in digital form, enabling people to experience it from anywhere in the world. He Sai is collaborating with a United Nations partner, TRIC, to create an immersive digital art piece for Du Fu Thatched Cottage in Chengdu. Du Fu Thatched Cottage is the former residence of Du Fu, a poet from the Tang Dynasty, who is revered as the "Sage of Poetry" in China.

Space Commerce: Navigate to products, not just physical stores

Through strategic cooperation with Gaode Map, a leading map and local discovery platform in China, Hesai is bringing immersive 3D environments into everyday location services. The reconstructed content of Gravity Art Museum and PIFO Gallery has been launched on Gaode Map's "Flying Street View" feature. Consumers can preview their destinations before visiting, while merchants can obtain updatable and analyzable digital spaces to optimize store layouts and product displays. With nearly 1 billion monthly active users on Gaode Map, this collaboration provides a vast platform for expanding the 3D experience. The ultimate goal is to extend navigation from "finding a store" to "finding a specific product inside the store."

Open ecosystem and two business models

Hesai is transforming Kosmo into an open spatial intelligence ecosystem, making raw collected data and 3D assets accessible, and integrating it with leading platforms such as NVIDIA Isaac Sim, Isaac Lab, Unreal Engine, RenderMan, and Houdini. This ecosystem supports two business models: integrated software and hardware solutions for robots, media, and other industry clients; and services for mass-producing high-quality real-world spatial data for world models and developers of Physical AI.

At the press conference, Hesai also announced the establishment of new cooperative relationships with Galbot and Dexmal to promote the development of embodied AI that utilizes Kosmo.

He Sai's Physical AI Route Map

Hesai is expanding its over a decade of experience in lidar, ASIC, and precision manufacturing to a broader Physical AI market. The company's lidar shipments have doubled for nine consecutive years, and it maintains a global leading position in the ADAS and robotaxi sectors of passenger vehicles. On this basis, the company is embarking on the next phase of growth around 3D world data and robotic motion.

"Physical AI is the infrastructure that enables machines to understand and act in the real world," said Li Yifan, co-founder of Hesai and also CEO, who added, "Our 3D world data engine and robot motion engine meet these two requirements respectively, and they are directly built upon Hesai's strengths, forming our second growth engine. Over the past decade, we have enabled machines to see the world; in the next decade, we will enable robots to understand and change the world."

In addition to Kosmo, Hesai is also expanding its robot actuator business, designing hardware around the application needs of the real world AI. Its actuator module powers the Wave dexterous hand of Sharpa, with a power density three times that of similar products and a 37% smaller volume. This dexterous hand has been incorporated into the reference design of the Isaac GR00T humanoid robot by NVIDIA and has been adopted by leading research institutions such as the University of California, Berkeley, and Stanford University. By the end of the second quarter of 2026, Hesai's actuator modules had accumulated shipments of over 10,000 units. The product portfolio is expanding from hand modules to various joints of the entire robot body, and it is expected that shipments will reach six digits in 2027.

Note:Source: Hesai Technology