LiPS: Large-Scale Humanoid Robot Reinforcement Learning with Parallel-Series Structures