Physical AI will revolutionize collaboration between humans and robots.

At the 'Next-Gen Human-Robot Collaboration with 5 Major Collaborative Robot Brands - Physical AI, Game-Changer Beyond Robot Limitations' seminar hosted by Safetics on the 11th at the Event Hall on the 3rd floor of POSCO Tower Yeoksam, Professor Seung-Ho Seo of the Department of Artificial Intelligence at Korea University gave a presentation on the topic of 'Executing AI, Collaborating Robots - The Current State and Industrial Applications of Physical AI'.

At this seminar, Professor Seo Seung-ho introduced the latest trends and real-world application examples of human-robot collaborative research utilizing generative AI and multimodal learning, drawing on his experience at the German AI Research Institute and industrial sites such as Samsung Electro-Mechanics.

He explained, “It is important how AI and robot technology connect with real life, and the series of cycles that extend from data collection and interpretation to the actual movements of robots.”br />
Professor Seo stated, “The digital world understands real life deeply enough for smartphones to predict our behavior,” and predicted that AI will evolve in a direction that has a more direct impact on daily life.

He emphasized the importance of human-centered robot design and control, stating, “Smart glasses and robots will become key devices driving changes in real life.”

'Physical AI' and 'VLA (Vision-Language-Action)' models are cited as recent trends attracting attention in the AI field.

Professor Seo stated, “The paradigm is shifting from existing perception-centered AI to physical AI that connects to actual physical behavior,” and introduced major examples such as Google DeepMind’s RT-2, SayCan, and OpenVLA.

These models are technologies that understand natural language commands based on large-scale multimodal data and translate them into actual robot actions.

Professor Seo explained, “On-device VLAs and cloud-based VLAs each have their own advantages and disadvantages, and the appropriate method will vary depending on the field environment.”

Professor Seo stated, “In industrial settings, the safety and flexibility of robots are paramount,” emphasizing that intuitive control via natural language and the ability to recognize and process various objects are essential.

It was assessed that the need for introducing humanoid robots is growing, particularly in Korean society, which is aging rapidly, across various sectors such as logistics and manufacturing.

In fact, the research team in which Professor Seo participates is currently working on a project to develop a foundation model-based humanoid robot to be deployed in logistics centers.

The goal of this robot is to recognize various boxes and objects and perform tasks according to natural language commands.

In addition, we are pursuing a quadruped robot project in cooperation with POSCO to assist with worker safety in high-temperature and high-risk environments.

Professor Seo stated, “Research combining wearable sensors and AI to accurately recognize workers’ actions is active.”

For example, we are developing technology that utilizes IMU and electromyography (EMG) sensors to precisely recognize even finger gestures and remotely control robots.

In addition, an experiment on safely controlling a robot in a digital twin environment was introduced.

To overcome the limitations of data collection, various attempts are being made, such as virtual data generation and text-to-motion conversion using generative AI.

Professor Seo said, “Reducing the gap between the real environment and virtual data, and generating natural motion that reflects physical constraints, is the challenge ahead.”

Professor Seo emphasized, “For AI and robot technologies to be practically applied in industrial settings, various technical and social challenges such as safety, cost, data standardization, and privacy must be resolved.”

In particular, he added that AI explainability, human-centered design, lightweight on-device AI, and industry-specific fine-tuning of foundation models are important.

Finally, he stated, “Cooperation across various fields is necessary so that AI and robotics can develop in a direction that improves the quality of human life,” adding, “I will continue human-centered AI and robot research in the future.”