On February 27, SoftBank and Ericsson successfully demonstrated Physical AI using a low-latency, highly reliable wireless network powered by AI-RAN. In this demonstration, AI analyzed camera and sensor data from robots to understand the environment and control robot movements.

In conventional robot development, the dominant approach has been onboard AI, where all AI processing is completed within the robot itself (e.g., Tesla, Boston Dynamics, Unitree, etc.).

The core of this demonstration lies in dynamic offloading technology, which allows robots to switch in real time—depending on the situation—between performing intensive computing locally on the robot or offloading it to external edge servers in the network. Based on the robot’s operational conditions and processing requirements, AI tasks that were previously executed solely on the robot can be dynamically offloaded to AI-RAN MEC.

With this approach, high-performance autonomous robots no longer require expensive GPUs on every unit. When advanced reasoning or complex path planning is needed, robots can leverage intelligence at the network edge (Edge AI). 

This significantly reduces hardware complexity and cost, enables lighter robot designs, and dramatically extends battery life. As a result, Physical AI can be deployed in industrial environments in a far more cost-efficient manner.

To enable this architecture, ultra-low latency, high-capacity, and highly reliable wireless connectivity between robots and edge AI is essential—naturally making 5G the ideal solution.

In this demonstration, SoftBank utilized its internally developed AI-RAN MEC. The low-latency, high-capacity, and highly reliable characteristics of the 5G network create the effect that the robot and external servers operate as if they were part of a single integrated system within the robot itself.

This architecture represents a shift from robots operating solely with onboard intelligence to connected robots that continuously become smarter through 5G networks and Edge AI. This is expected to make the large-scale deployment of affordable, high-performance robots a practical reality.