Trajectory planning of quadrotor elastic suspension system based on DDPG algorithm

Most of current research on the system of a UAV carrying suspended loads only considers inelastic ropes. Many traditional algorithms have been developed to keep the load's swing angle and the path lengths within some reasonable range, which can result in longer path lengths and larger load swing angles. To tackle these challenges, we establish a three-dimensional dynamic model of the elastic suspension system for quadrotors and design a reward function for DDPG. Additionally, LSTM networks and a prioritized experience replay mechanism are utilized. The system is trained and compared with traditional algorithms through simulations in two complex environments. Finally, the effectiveness of this method is demonstrated through real time flight experiments.