The primary methods used to generate motions for virtual environment characters include Procedural Animation, finite-state machines (FSM), and Behavior Tree techniques. However, these traditional methods are labor-intensive, requiring manual specification of motion animations frame by frame or generating motions via sensor-based motion capture. They also face limitations in terms of flexibility and scalability. To address these challenges, reinforcement learning-based motion generation methods have recently gained attention. In this study, we propose an environment where a humanoid character, based on the SMPL model, can learn and perform various motions using the PPO reinforcement learning algorithm. Specifically, we investigate the flexibility and potential of automated motion generation by applying the SMPL model to a physics-based humanoid in the ML-Agents platform.