Professor Lee Yoon-sang from the School of Computer Science at Hanyang University and his research team drew major attention at SIGGRAPH 2025, the world’s premier international conference on computer graphics, by presenting their PhysicsFC technology. The team presented their findings in the Technical Papers program and was also invited to demonstrate their work in the Emerging Technologies program. Their technology attracted particular interest from developers of the world-famous basketball game series NBA 2K during the conference.

The ‘PhysicsFC’ technology is based on deep reinforcement learning and enables the control of characters simulated using a physical engine, so they can move and handle the ball like real soccer players. Professor Lee began this research with the belief that advancements in AI would allow characters in simulation games based on physics to perform actions similar to human players. He reasoned that integrating such physics-based control into soccer games would result in more realistic motion of characters for users.

Traditional soccer games often use kinematic motion that does not account for physical interactions, leading to unnatural character movements. To address this problem, Professor Lee designed an effective reward system for control policies learning for each skill. He also carefully established the initialization and termination conditions for each ‘episode,’ which represents a single trial unit in reinforcement learning.

‘PhysicsFC’ integrates essential skills such as kicking, moving, dribbling, and trapping within a finite-state-machine-based control structure. Among these, trapping, meaning receiving a ball coming through the air, is a key action in real soccer that previous studies had not addressed.

The research team proposed a new method called ‘STI(Skill Transition-Based Initialization)’ to enhance character agility and ensure smooth transitions between actions. When each skill is trained independently, transitions between motions can appear unnatural. To tackle this, they set the initial state for training the next skill as the positions of the character and ball from the previous skill.

They also considered more than just target positions during reward computation and addressed the issue of unnatural character movements by introducing the ‘DEGCL(Data Embedded Goal-Conditioned Latent Guidance) method,’ which takes into account target figures embedded in motion capture data. Motion capture data refers to time-series posture data obtained by measuring real human movements using sensors or cameras.

Professor Lee did not start his career in computer graphics. Initially preparing for graduate studies in mechanical engineering, his undergraduate major, he eventually found himself drawn to computer graphics, a field he explored passionately through club activities. Ultimately, he chose to pursue what he truly enjoyed. “Since this wasn’t my original field, it was challenging at first,” he said, “but doing what I love made research genuinely enjoyable.” After completing his doctoral studies and spending some time in industry, Professor Lee returned to academia with the desire to tackle challenging research problems.

Offering advice to Hanyang students concerned about their future paths, he said, “You need to know who you are and find something you truly enjoy—something that fits you.” He added, “I altered the direction of my life by asking myself what kind of work I could enjoy doing, and what would make me feel both happy and fulfilled.” He emphasized, “Don’t focus too much on what others say or what you hear around you. Instead, listen to your inner voice and take time to reflect on yourself.”

Since the beginning of his appointment at Hanyang University, Professor Lee has pursued his research on graphics, considering branching out to robotics. He continues to expand his work in physics-engine-토토사이트 로그인 없이 character control into robotics research, exploring new directions and innovative applications 토토사이트 로그인 없이 on the control technologies he has developed.