Kang, KyungwonGu, TaehongKwon, TaesooZordan, Victor2024-08-202024-08-202024978-3-03868-263-9https://doi.org/10.2312/sca.20241165https://diglib.eg.org/handle/10.2312/sca20241165We propose a physics-based climbing controller that consists of two learning stages. Firstly, a hanging policy is trained to grasp holds in a natural posture. Once the policy is obtained, it is used to extract the positions of the holds, postures, and grip states, thus forming a dataset of favorable hanging poses. Subsequently, a climbing policy is trained to execute actual climbing maneuvers using this hanging state dataset. The climbing policy allows the character to move to the target location using limbs more evenly. Experiments have shown that the proposed method can effectively explore the space of good postures for climbing.Attribution 4.0 International LicenseCCS Concepts: Computing methodologies → Physical simulation; Motion capture; Reinforcement learningComputing methodologies → Physical simulationMotion captureReinforcement learning10.2312/sca.202411652 pages