Physical AI in Retail: Applications in Logistics, Stores, and Foodservice

Artificial intelligence writes texts, analyzes data, and generates images. Physical AI goes one step further: It refers to AI systems that can perceive their real-world environment, process information from it, and respond through physical systems. This allows machines to interpret situations and derive actions from them—from autonomous transportation in warehouses to capturing inventory levels in stores.

The term “Physical AI” has been strongly popularized in recent years, among others, by NVIDIA CEO Jensen Huang. At CES in January 2025, he highlighted the connection between artificial intelligence, robotics, and autonomous machines and said that the “ChatGPT moment for general robotics” was just around the corner. He was referring to a developmental leap similar to the one ChatGPT triggered for generative AI. At the same time, NVIDIA introduced Cosmos, a platform for developing Physical AI systems. On January 5, 2026, at CES, Huang finally declared that the “ChatGPT moment for robotics” had arrived.

Traditional robots are often designed for clearly defined, repetitive processes. For example, they may perform the same movement in the same place every time. Physical AI adds an intelligent layer to such machines: Sensors detect changes, AI interprets the situation, and the system adjusts its actions accordingly.

That is why not every robot is automatically Physical AI. Conversely, Physical AI does not necessarily have to look like a robot. Autonomous vehicles and other machines can also fall into this category. Physical AI and robotics overlap, but they are not the same thing.

Computer vision, sensor technology, and digital twins are also not Physical AI in themselves. However, they can be important building blocks for it.

Digital twins are digital representations of real-world systems, spaces, or machines that model their conditions and behavior and can be used for simulations. In these simulation environments, processes can be tested and AI models trained before a system is deployed in the real world.

This is particularly relevant for Physical AI: A robot or vehicle can first test situations virtually. New configurations and software changes can be tested without immediately interfering with ongoing physical processes.

The potential of Physical AI becomes particularly clear wherever digital decisions are implemented directly in the real world. This mainly concerns processes in which goods need to be moved, inventory needs to be recorded, or workflows need to adapt flexibly to changing conditions.

The following examples show the wide range of ways Physical AI and related autonomous systems can be used in retail. They range from applications in logistics and intralogistics to inventory and shelf scanning in stores and retail foodservice. Technologies used include coordinated robotic systems, autonomous vehicles, and scanning and cooking robots.

Several current examples illustrate how different these applications can look.

The Otto Group is working with NVIDIA and Reply on a “Robotic Coordination Layer” for logistics. At the Hermes Fulfilment logistics center in Löhne, a digital twin provides a virtual representation of the facility, including the robotics used there.

According to the Otto Group, different types of robots are to be simulated, integrated, and coordinated through this system. New processes can first be tested virtually. In the future, robot fleets, sensors, warehouse management systems, and conventional automation technology are expected to work together more closely as a connected system.

In 2025, Amazon tested Blue Jay, a system that coordinated multiple robotic arms for different work steps. Since February 2026, Amazon has no longer used Blue Jay itself in operational settings; however, the technology developed for it is expected to continue to be used.

Belgian retail group Colruyt Group is using autonomous systems in logistics. Together with KION, a provider of warehouse technology, industrial trucks, and automation solutions, the company is further developing self-driving pallet vehicles. According to the companies, artificial intelligence and computer vision are intended to help the vehicles perceive their surroundings, detect obstacles, and move independently through distribution centers.

This example shows that Physical AI does not have to take the form of a traditional robot and can also be used in autonomous vehicles.

Autonomous systems that use individual components of Physical AI are also being tested on the sales floor.

In 2026, SPAR Austria expanded its trial of the Tally inventory robot to additional stores. The robot moves independently through the aisles and scans shelves. It detects inventory and gaps and transmits the information it collects to employees.

At dm, autonomous scanning robots are also in use. They scan shelves and use the data to create digital representations of the stores. The data can be used, for example, to analyze product arrangements or optimize routes for click-and-collect orders.

At the same time, these examples show that autonomous robotics is not automatically the same as Physical AI. Systems can move independently, perceive their surroundings, and provide data for additional processes without necessarily demonstrating the complete cycle of a Physical AI application. The decisive factor is whether the captured information is evaluated using AI and whether the system independently derives actions from it in the physical world.

The connection between software and physical action becomes particularly clear in retail foodservice.

At REWE in Düsseldorf-Heerdt, the CA-1 cooking robot from Circus went into operation at the end of 2025. The system combines robotics, artificial intelligence, and food technology. Once an order is placed, it largely automates the cooking process, from the prepared ingredients to the finished dish.

Physical AI becomes directly visible here: AI does not remain in the background of a digital process but instead controls a machine that performs a specific task in the real world.

For retail companies, Physical AI is particularly relevant wherever processes are difficult to define entirely through rigid rules. Warehouses change, people move through sales areas, and products are not always positioned exactly where they are supposed to be.

Physical AI is intended to enable machines to deal with such changes more flexibly. This gradually shifts automation away from fixed, preprogrammed processes toward systems that can perceive their environment and adapt their behavior accordingly.

However, current applications also show that the fully autonomous retail world is not yet a reality. Many solutions are pilot projects or are specialized for clearly defined tasks.

What is Physical AI in simple terms?
Physical AI combines artificial intelligence with physical systems. Machines can perceive their environment through sensors, interpret information, and derive physical actions from it.

Is every robot Physical AI?
No. A fully preprogrammed robot can also perform tasks automatically. Physical AI adds the ability to perceive the environment and adapt its own actions accordingly.

Where is Physical AI used in retail?
Physical AI and related autonomous systems can perform a variety of tasks in retail. The examples presented in this article range from logistics and intralogistics to autonomous transportation and robotic systems, inventory and shelf scanning, and retail foodservice.

Are digital twins Physical AI?
No. A digital twin is initially a virtual representation of a physical system. However, it can provide an important foundation for Physical AI because machines and AI models can be trained, tested, and validated within it.