Reinforcement Learning-based Jump Control Method of Tethered Quadruped Robots for Lunar Lava Tube Exploration

As an important celestial body in the solar system, the moon has lunar lava tubes on its surface that provide natural protective barriers, making them ideal locations for lunar base construction and scientific research missions. However, the complex terrain and low visibility within the lunar lava tubes make it difficult for existing wheeled robots to efficiently perform exploration tasks. Therefore, this study proposes a solution using a tethered quadruped robot for exploration, combined with deep reinforcement learning control methods to enhance its jump control and posture adjustment capabilities within the lava tubes. This study develops a hybrid dynamic model that includes both rigid and flexible bodies, and provides a detailed analysis of the coupling dynamics between the quadruped robot and the tether system. Through deep reinforcement learning algorithms, the posture control and jump efficiency of the quadruped robot during takeoff and landing are optimized.