RTSensor Data-Driven Modular Network Model for Jump Motion in One-Legged Robot

In this study, we propose a three-phase learning network-based reinforcement-learning model to implement the jumping motion of a one-legged robot. The jumping motion of a one-legged robot is divided into three phases, each of which can be trained independently. This method ensures high performance and faster convergence through partial optimization and efficient learning. Furthermore, to mitigate the discontinuities during phase transitions, we propose a switch-transition algorithm. The results indicate that the switch-transition model complements the movement during the transition section, thus ensuring continuity throughout the entire jumping motion.