The Korea Advanced Institute of Science and Technology (KAIST) is joining hands with Nvidia to develop physical artificial intelligence (AI) that understands human movement and intent. The plan is to build a general-purpose AI model that predicts a person's next movement by learning motion data collected from real people and wearable robots, and to apply it to humanoids, rehabilitation robots, and digital healthcare.
KAIST announced on the 26th that it will establish the 'Human Physical Intelligence Technology Center' for joint research with Nvidia in the physical AI field. The center will be operated based on KAIST's Human Physical Intelligence Research Center in the Department of Mechanical Engineering, with Professor Kong Kyoung-chul, an expert in wearable robots, and Professor Kim Jung, an expert in bio-robots, serving as co-directors. While the joint research institute announced by the two sides on the 24th researches core agentic AI technologies specialized for the Korean language and domestic industries, this center focuses on developing physical AI technology that learns human motion data and applies it to wearable robots and humanoids.
The main goal of the joint research is to build a 'Human Motion Foundation Model.' It is a foundation model that understands, predicts, and generates various movements by training AI on large-scale data covering human walking, joint movements, muscle force, and the process of maintaining balance. While existing robots moved according to pre-programmed movements and situation-specific control programs, robots applying this model can analyze the user's posture and movement to predict the next action. A wearable robot could add appropriate force to the necessary joints at the moment the user takes a step, or a humanoid could adjust its work speed and movement paths to match a person's movements.
KAIST has accumulated human motion data and motor control technology while developing wearable robots for walking assistance and rehabilitation and demonstrating them with actual users. Professor Kong also has experience commercializing wearable robots through his startup Angel Robotics. Unlike generative AI, which can learn from sentences and images accumulated on the internet, physical AI requires data generated as people move and exert force in real environments. This is why KAIST's real motion data is considered a key asset in this collaboration.
From Nvidia, Charles Cheung, senior manager at the Nvidia AI Technology Center, will participate to provide technical advice and development support. Research and education programs utilizing Nvidia's virtual-space building platform 'Omniverse' and digital twin technology will also be operated. The two sides plan to expand the scope of research in the future from wearable robots and humanoids to manufacturing, shipbuilding, steel, logistics, and healthcare.
The joint center will review its research goals and achievements every six months and hold an international symposium inviting researchers from home and abroad each year. It will also run a student ambassador program that selects 5 to 10 KAIST students to participate in lectures, mentoring, and joint projects with Nvidia experts.
