SMPL-based Character Walking Motion Generation Using ML-Agents

Techniques such as procedural animation, finite-state machines (FSM), and behavior trees have generally been used to generate character motions in existing games or virtual environments such as metaverses. However, these methods are labor-intensive because motion animations must be manually created frame by frame or motions must be generated through sensor-based motion capture. Motion generation methods using reinforcement learning have recently attracted attention as an alternative address this limitation. In this study, we discuss a method for generating walking motions by replacing the simple skeleton-based agent used in reinforcement learning with an SMPL-X model.