---
title: "Hesai Sells a Camera to Teach Robots Where Things Are. An Ice-Cream Stand Shows Why."
date: 2026-11-20
category: Humanoids & Robotics
site: NeuroAI
canonical: https://neuroai.site/a/na-robot-hesai-kosmo-spatial-data
language: en
---

# Hesai Sells a Camera to Teach Robots Where Things Are. An Ice-Cream Stand Shows Why.

> A major Chinese lidar maker has launched a mapping camera for robot-training data, betting the next big hardware market is not the robot itself but the sensors and actuators inside it.

A robot hands a paper cone across a counter in a Shanghai mall. It does not have a camera perched on its head, and nobody has spent a week teaching it the layout of that particular shop. What it has is a nearby device that has already recorded the room in three dimensions.

The ice cream is the easy part. The harder question is how a machine ever learns enough about an ordinary place to work in one it has never seen before.

## What Hesai Actually Launched

On 14 October 2026, Hesai Group (2525.HK), a Chinese autonomous-driving technology company, unveiled a device called Kosmo at its Shanghai factory. It combines lidar (激光雷达) sensors with conventional cameras to produce detailed three-dimensional maps of real spaces.

Hesai is not a small player trying to invent a new identity. Over the past three years it has become a leading supplier of lidar for autonomous vehicles, and says it has pushed the unit cost of that technology below US$200 (≈ HK$1,560). Its customers include Mercedes-Benz, Volkswagen, General Motors and Ford. The car business is mature; the company wants a next place for its sensors.

That place, in the view of co-founder and chief executive David Li, is humanoid (人形机器人) robotics. At the launch, Li described a gap in the high-quality spatial data needed to train the AI systems that let robots understand and move through the real world. Most robots, he said, are still confined to low-impact functions like greeting guests, waving or dancing.

Kosmo is pitched as a way to fill that gap. Developers point it at a space, capture the geometry, and hand the resulting data to a robot-training pipeline. Hesai says it has signed agreements with Chinese robotics firms to use the product, but Li declined to give revenue projections for it.

## Why a Map Is Worth More Than a Demo

Anyone who has watched a robot video online knows the problem. A machine that can run, jump or do a backflip is impressive for about thirty seconds. A machine that can reliably put an object on a shelf, in a building it has never entered, under lighting that changes at 4pm, is the thing factories actually want to buy.

The gap between the two is data about physical space. Text-trained chatbots can scrape the internet; robots cannot. They need consistent measurements of rooms, objects, distances and surfaces.

This is why a lidar company sees an opening: handheld sensors collect spatial data faster than asking people to film the same task a hundred times. Hesai's own demonstration partner makes the point concrete. Sharpa Robotics, a startup founded by Hesai's co-founders, uses a robotic hand called Sharpa Wave that has been serving ice cream at a Dairy Queen outlet in Shanghai. Hesai says Kosmo's mapping helped the robot reach a task success rate of 99 percent.

Treat that number as a company claim, not an independent result. The test conditions, the definition of a "task", and the period measured were not disclosed. But the example is still useful: a service job in a real shop, rather than a stage routine, is exactly the kind of work the industry is trying to make routine.

## The Parts Nobody Photographs

Kosmo is only half of Hesai's robot pitch. The company is also developing robot actuators (执行器) — the components that turn electricity into mechanical motion. Li said actuators account for more than half of a robot's material costs, and that demand is growing for versions that are lighter, smaller and more precise so machines can be safer around people.

This is a less flashy strategy than building a humanoid, and easier to evaluate. A robot company needs a brain, a body and a supply chain; Hesai wants to supply the eyes and muscles, as it once supplied lidar to carmakers. Li explicitly compared the ambition to Bosch's role in the automotive industry — a tier-one supplier that sells to many manufacturers rather than competing with them.

Sharpa is the proving ground. It operates independently of Hesai, but uses technology from its founders' former employer in its flagship hand. The startup recently raised more than 4.5 billion yuan (≈ US$672 million / HK$5.24 billion) in a funding round, according to Reuters. That figure is large enough to signal that investors are willing to back component suppliers as well as complete robots.

## What Still Needs Checking

Three things are missing from the public picture.

First, the 99 percent figure. Without a published protocol, it is impossible to tell whether it describes a controlled demonstration or a week of real service. Second, the economics. Hesai declined to provide revenue projections for Kosmo, so there is no public evidence of what a unit costs or how quickly it pays for itself. Third, the customer list: "agreements with Chinese robotics firms" is not the same as named deployments with volume commitments.

Better maps may improve today's robots but do not solve reasoning: a machine that knows where a cup is still has to decide what to do when a customer moves it. Spatial data is a foundation, not a brain.

## Key takeaways

- Hesai, a leading automotive lidar supplier, launched Kosmo, a mapping camera that combines lidar and conventional cameras.

- The product targets a shortage of real-world spatial data for training humanoid (人形机器人) systems.

- A Sharpa robotics hand serving ice cream in Shanghai is the reference case; Hesai claims a 99 percent task success rate.

- Hesai is also building actuators (执行器), which it says make up more than half of a robot's material costs.

- No revenue projections or test methodology were disclosed.

## Honest limitations

The 99 percent success rate and the cost comparison for actuators both come from Hesai and its CEO. I could not find an independent evaluation of either. The ice-cream robot is a real deployment, but a single site in a controlled retail setting is not evidence of general reliability.

The 4.5 billion yuan (≈ US$672 million / HK$5.24 billion) funding figure for Sharpa is reported by Reuters; the company's valuation and the exact close date were not part of that report. The US$200 lidar cost figure is likewise Hesai's own statement about its automotive business, not a market-wide price survey.

A separate Reuters story in November describes Chinese robot makers shifting investment from hardware toward robot AI; this piece is about one supplier's launch, not the whole sector.

Exchange-rate conversions use the rate cited in the Reuters report (US$1 = RMB 6.6923) and a rounded Hong Kong dollar peg. Actual transaction values will differ.

## What readers can do now

- **If you are evaluating a robot vendor, ask how its training data was collected.** A flashy demonstration with pre-mapped surroundings says little about performance in a new building.

- **If you invest in robotics, watch component suppliers as closely as robot makers.** Actuators and sensors are where a large share of the bill of materials sits, and they are easier to compare across companies.

- **If you run a retail or logistics site, treat success-rate claims as a starting question, not an answer.** Ask what the failure cases looked like before you sign anything.

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-robot-hesai-kosmo-spatial-data
Free to quote with attribution and a link to the original.
