The center of gravity of AI technology is shifting beyond cloud-based software competition to 'Physical AI,' which designs and verifies real-world physical behavior.

The 'e4ds Physical AI Frontier 2026' conference, which surveys these trends of change, will be held on March 6 at Rooms 317 and 318 of COEX in Samseong-dong, Seoul.

This event is the largest specialized physical AI conference in Korea, focusing on the implementation of AI that actually operates in industrial settings, such as sensors, semiconductors, embedded systems, robots, and manufacturing automation.

It redefines AI from a systems engineering perspective, encompassing everything from sensor signal quality to on-device inference, real-time control, and verification of power, latency, and stability, rather than a simple competition in algorithm performance.

Physical AI refers to an integrated architecture where sensors recognize physical signals, chips such as MCUs, NPUs, and APUs infer the data locally in real-time, and then verify the operation of actuators and motors through a control loop.

In this process, signal matching, noThe balance between product execution variables such as latency suppression, memory footprint, latency, power budget, and interface compatibility acts as quality competitiveness.

In the keynote speech, Lee Soon-youl, President of the Korea Industry 4.0 Association, presents implications for practical application in industrial settings, while Matteo Maravita, Center Director at STMicroelectronics, discusses the potential for implementing embedded AI and sustainable generative AI.

The afternoon event is divided into Track 1 and Track 2.

Track 1 focuses on strategies for applying physical AI in industrial settings, market trends, safety technologies, and case studies of collaborative robots and manufacturing automation.

Choi Hong-seop, CEO of Maum AI, will present the physical AI value chain and South Korea's strategy, while Kim Hwi-yeon, CSO of Safetics, will discuss technical responses to ensure the safety of robots and physical AI. Dobot Robotics CEO Lee Nae-hyeong introduces a new paradigm of manufacturing automation created by collaborative robots and physical AI.

Track 2 covers key elements of technology implementation, such as embedded semiconductors, on-device AI, simulation-based reinforcement learning, model lightweighting, and high-speed interconnects.

ETRI Division Head Jeong Young-jun surveys the present and future of physical AI through examples of quadruped robots and wearable suits, while Qualitas Semiconductor Executive Director Kim Jae-young emphasizes the importance of high-speed interface technologies such as UCIe, PCIe, and MIPI.

An official from e4ds stated, “This conference presents a ‘methodology for completing physical AI into products and services’ rather than treating it as a research concept,” adding, “It will serve as a practical benchmark for developers and companies aiming to make AI work in reality.”