Hesai Kosmo Spatial Camera Explained: Why 3D Vision Could Accelerate Humanoid Robotics

Quick answer: Hesai’s new Kosmo spatial camera is designed to capture detailed 3D information about the physical world for AI-powered robots. By combining lidar and camera data, systems like Kosmo can help humanoid robots perceive geometry, objects and movement more accurately — a key requirement for moving robotics from demonstrations into reliable real-world work.

Key takeaways

• Kosmo combines lidar and imaging to capture rich spatial data.
• Robotics needs high-quality physical-world data, not only text and images from the internet.
• Better perception can improve navigation, manipulation and safety.
• Spatial cameras may become a critical infrastructure layer for training embodied AI.
• The long-term opportunity is bigger than one sensor: robotics will need integrated perception, actuation, planning and learning systems.

What is Hesai Kosmo?

Kosmo is a spatial camera from Hesai, a company best known for lidar technology used in autonomous driving. Reuters reported that the device combines lidar sensors and cameras to capture three-dimensional information for robotics applications.

The aim is to provide machines with a richer representation of the environment than a conventional camera alone.

Why robots need spatial intelligence

Language models can learn from enormous text datasets, but a robot must understand distance, depth, surfaces, object position and physical constraints. A cup is not simply a visual pattern. A robot needs to know where the cup is, how far away it is, how to approach it and how much force to apply.

Spatial sensors help convert the real world into machine-readable geometry.

Lidar plus cameras

Cameras are excellent at capturing texture, color and semantic detail. Lidar is strong at measuring distance and 3D structure. Combining both can give a robot complementary information: what something appears to be and where it exists in space.

Sensor fusion is therefore central to robotics. A system can cross-check multiple signals rather than relying on one imperfect source.

Why data collection may be the hidden bottleneck

Humanoid robotics has improved quickly, but reliable physical intelligence requires enormous amounts of training data. Internet-scale text is abundant. High-quality data showing physical interactions in diverse environments is much harder to collect.

Spatial cameras can help create datasets that capture rooms, objects, movement and manipulation. That data can be used to train perception systems and potentially world models that predict how physical environments change after an action.

From perception to manipulation

Seeing the world is only the first step. A useful robot must turn perception into action. That requires planning, control systems and actuators that can move with precision.

Reuters also reported that Hesai is developing compact actuators, suggesting the company wants to supply more of the robotics stack rather than only sensing hardware.

Why humanoid robots are difficult

Human environments are unstructured. Doors vary. Floors are uneven. Objects move. People behave unpredictably. Tasks that feel trivial to humans — picking up a slippery object or opening a drawer — require continuous perception and correction from a robot.

This is why impressive demonstrations do not automatically translate into reliable commercial deployment. Real-world robotics needs extremely low error rates across thousands of edge cases.

What Kosmo could enable

Potential applications include warehouse automation, manufacturing, service robotics, research data collection and training humanoid systems. Spatial cameras can also support digital twins, mapping and simulation pipelines.

The strongest value may come when physical data is captured repeatedly at scale and fed back into model training.

Privacy and safety

Robots that continuously map environments can collect sensitive spatial and visual information. Companies will need clear rules about storage, retention, consent and where processing occurs.

Safety is equally important. A perception error in a chatbot creates a bad answer; a perception error in a physical robot can damage property or injure someone. Robotics therefore requires stronger validation than many purely digital AI applications.

VIG take

The race to build humanoid robots will not be won by language models alone. It requires a full stack: sensors, spatial data, world models, planning, actuators and safety systems. Kosmo is interesting because it targets one of the least glamorous but most important layers — giving machines better data about physical reality.

FAQ

What does a spatial camera do?

It captures information about the three-dimensional structure of an environment, helping machines understand depth, distance and object position.

Why use lidar with cameras?

Cameras provide rich visual detail while lidar provides precise distance and geometry. Combining them can improve perception reliability.

Is Kosmo only for humanoid robots?

No. Spatial sensing can be useful across robotics, mapping, autonomous systems and simulation workflows.

Source

• Reuters: Hesai launches Kosmo spatial camera

Disclosure: Independent editorial analysis by Vector Insight Global. Not sponsored by Hesai.