Coordinated Control of Flight and Morphing for Morphing Quadrotor via Reinforcement Learning

This paper proposes a reinforcement learning (RL) approach to coordinated control of flight and morphing for morphing quadrotors. A multi-expert weighted combination-based flight controller (MEWC-FC) is constructed, which generates morphing-specific flight control actions from a group of expert control actions. An extreme learning machine and particle swarm optimization (ELM-PSO) algorithm is developed for pre-training, i.e., the PSO algorithm optimizes the experts, and the ELM network initializes the morphology-related combined weights. Then, the ELM network is fine-tuned under RL framework to improve the generation performance. Besides, to achieve coordination with MEWC-FC to further optimize flight performance, an RL-based flight state-related morphing controller is designed. The MEWC-FC and the morphing quadrotor dynamics are set up as the training environment in the framework of RL to investigate the coupling relationship between them. Finally, simulation results show that the proposed ELM-PSO with RL tuning algorithm can produce the MEWC-FC with generalization capabilities, including zero-shot generalization to unseen morphologies. The MEWC-FC can realize the morphing-specific flight control over the whole morphing space. Moreover, the coordinated control of flight and morphing exhibits better performance compared to the fixed-structure flight.