Hesai CEO: A Robot in a DQ Ice Cream Shop Needs More Than Hands and a Model — It Needs to Understand Space

Handing an ice cream shop over to a robot takes a manipulator, a dexterous hand, cameras, and a model that perceives the scene and makes decisions. All of this is visible in the demo. But according to Hesai co-founder and CEO Li Yifan, the list does not end there: a number of tasks tied to space remain. He said this at a presentation while discussing the collaboration between Sharpa and DQ. The Chinese outlet Jiqizhixin put a different part of the story in its headline: the real world can be brought into AI in 15 minutes with one camera, and, as the outlet writes, Fei-Fei Li would appreciate that.

An important caveat: the available excerpt of the article does not describe the solution itself. It gives no product name and no data on accuracy, cost, or hardware. Below is what the text supports, and how this pairing can reasonably be read.

What we know about the case

There are three participants. Hesai is known as a lidar maker, so a talk about space from its CEO sounds concrete: the company lives in the world of 3D data. In this story, Sharpa handles manipulation, meaning the mechanics and dexterity. DQ supplies the scenario: an ice cream point of sale, where the robot has to do specific work alongside people.

The retelling shows how Li Yifan builds his argument. First comes the "visible" stack: the arm, the hand, the sensors, the model. Then he moves on to what stays off-screen. A robot in a shop has to know where things are, how the area around it is laid out, what has changed since the last shift, and where it can move without risk. This is knowledge about a specific room, not about the world in general.

Why Fei-Fei Li is mentioned

For many years, Fei-Fei Li has pushed the idea of spatial intelligence: AI should understand a 3D scene, not just text and flat images. The reference in the headline shows how the Chinese press reads Hesai's talk. If a real scene can be digitized quickly and cheaply, robots get a model of the place where they will work.

This also explains "15 minutes, one camera." For commercial robot deployment, what matters is not only manipulation quality but also the time it takes to prepare a new site. If every coffee shop or store takes weeks of manual setup, scaling will not work. If a digital copy is ready in a quarter of an hour, the economics look different. The text does not confirm these figures; they are a claim from the headline.

What the article records

• Speaker: Li Yifan, co-founder and CEO of Hesai.
• Partners in the case: Sharpa (manipulation) and DQ (application scenario).
• Scenario: an ice cream shop, where the robot needs to work with a real physical space.
• Claimed scene digitization time: 15 minutes.
• Claimed capture hardware: one camera.
• Not disclosed: product name, reconstruction accuracy, data format, price, and release timeline.

How to read this in practice

For teams building robots for service and retail, the takeaway is simple: deployment cost is increasingly determined by preparing the environment, not by the power of the arm. When you see a robot demo, ask three questions. How long does it take to prepare a new location? What happens if furniture or equipment is moved? Which sensors are needed at the site itself, and which only during the initial scan?

For the market, something else matters more. Hesai, used to selling lidars, is entering the conversation about spatial data for embodied AI, and that widens its role in the supply chain. For now, this is a position stated at a presentation, not a verified product. It is worth waiting for technical details and independent tests before treating 15 minutes and one camera as a working standard.