SK Telecom and KT develop AI brains for robots

Automatically extracted; no manual editorial approval recorded.

application scope. Recently, global tech and telecom analyst firm Technexsus released a report analyzing the physical AI business status of 40 global telecom companies. Among them, six were highlighted as leaders for combining computational infrastructure, communication networks, and collaborations with robotics firms. South Korean telecoms SK Telecom and KT were included, alongside NTT Docomo and SoftBank (Japan), D

Open source S3

eutsche Telekom (Germany), and T-Mobile (U.S.). ◇Telecoms Shift from Connecting Robots to Enabling Learning Previously, telecom companies focused on linking robots to communication networks or supplying serving and delivery robots. However, with AI advancements, robots now require technologies to understand their surroundings, adapt to new tasks, and learn autonomously—beyond repetitive operations. The importance of

Open source S4

managing multiple robots collaboratively has also grown. As computational demands for robot learning and operation rise, telecoms’ data centers, GPUs, and networks present new opportunities. The process of AI robots learning tasks and operating in real-world settings reveals telecoms’ roles at every stage. First, robots must be trained via AI to perform tasks and tested for accuracy. This requires processing large da

Open source S5

tasets—such as camera footage, sensor values, and motion data. Computational resources to simulate and test scenarios in virtual environments are also essential. Telecoms, which have provided these resources through data centers and clouds, are leveraging their strengths. For instance, Deutsche Telekom partnered with NVIDIA to establish an ‘industrial AI cloud’ in Munich, Germany, which began operations in February.

Open source S6

It allows robotics firms to borrow computational resources for AI development and test robot tasks in virtual environments. SK Telecom is preparing a robot learning platform based on digital twin technology for manufacturing, developed with SK Hynix. SoftBank also built a development tool that collects robot motion data via GPU cloud, trains AI models, and applies them to real robots after virtual testing, conducting

Open source S7

a pilot with Yaskawa Electric in July. ◇Assisting On-Site Judgment and Managing Multiple Robots… Developing ‘Brains’ Even after training, robots require computation during on-site operations. This includes ‘AI inference’—identifying objects in camera feeds and deciding actions like grasping or movement directions. While robots can process this via internal semiconductors, offloading complex tasks to external servers

Open source S8

via networks is being explored to avoid limitations from high power consumption, heat, and weight. T-Mobile announced on October 27 that it is testing a method with humanoid robot firm Figure AI to handle part of a robot’s AI computations on nearby external servers. This extends telecoms’ role from network connectivity to supporting robots’ decision-making processes. Deploying multiple robots in the field also requi

Open source S9

res systems to monitor their status and allocate tasks. This involves integrating robots from different manufacturers and synchronizing them with existing factory equipment. KT is utilizing its ‘K-RaaS’ platform, which manages diverse robots and machinery, for physical AI demonstrations. Robots, a subsidiary of NTT Docomo, is developing integrated control technology linking walking robots, robotic arms, and sensors,

Open source S10