‘FreeMusco(Motion-Free Learning of Latent Control for Morphology-Adaptive Locomotion in Musculoskeletal Characters)’를 개발했다고 밝혔다. Hanyang University(Lee Ki-jeong as President) has announced that Professor Lee Yoon-sang’s research team from the Department of Computer Science and Engineering developed a new artificial intelligence technology called ‘FreeMusco(Motion-Free Learning of Latent Control for Morphology-Adaptive Locomotion in Musculoskeletal Characters).’ The technology allows musculoskeletal-based characters to learn natural movement without the need for motion capture data.

Conventional research on physics-based character control has relied heavily on motion capture data—recordings of real human movements. While effective for realistically reproducing human actions, this method has limitations when applied to animals or imaginary creatures where collecting motion data is difficult or impossible.

To address this, the research team proposed the ‘FreeMusco’ framework, a ‘Motion-Free’ learning system. This technology utilizes musculoskeletal simulations, which precisely model muscle and bone structures, as physical ‘prior knowledge.’ It is designed so that energy-efficient and stable walking behaviors emerge naturally based on physical constraints and biomechanical structures, without requiring separate demonstration videos or motion data. A key feature is that different movement strategies appear spontaneously depending on the character's morphology.

As a result, the team confirmed that energy-efficient movements suitable for each shape were automatically formed: while humanoid characters maintained bipedal walking, a virtual character with modified legs, 'Chimanoid,' independently chose quadrupedal walking.

Furthermore, by learning the latent space alongside the control policy, the research team demonstrated that the learned walking controls could be expanded to high-level tasks such as goal navigation or path following. This marks the first case of musculoskeletal character control technology learned without 토토사이트 토토나라 data being applied to such tasks.

Professor Lee Yoon-sang stated, “FreeMusco is a new approach where movement is formed on its own through physical laws and biomechanics, rather than simply ‘mimicking a correct answer’ motion. We expect this to expand into various fields, including animal simulation, virtual creatures in games and movies, and robot control.”

This research is expected to suggest a new direction for next-generation physics-based character animation and robotics by presenting a universal loco토토사이트 토토나라 control technology applicable even to characters for whom obtaining 토토사이트 토토나라 capture data is impossible.

This study was conducted with support from the National Research Foundation of Korea(NRF) and the Korea Creative Content Agency(KOCCA). The results were selected and presented at 'SIGGRAPH Asia 2025,' a world-renowned academic conference in computer graphics and interactive techniques held in Hong Kong from December 15 to 18.

In the paper, FreeMusco: 토토사이트 토토나라-Free Learning of Latent Control for Morphology-Adaptive Loco토토사이트 토토나라 in Musculoskeletal Characters, Hanyang University doctoral student Kim Min-kwan participated as the first author, and Professor Lee Yoon-sang participated as the corresponding author.