The invention provides an armed unmanned aerial vehicle hovering stability control method based on model-free incremental
reinforcement learning, and the method comprises the steps: building an unmanned aerial vehicle and on-load mechanical arm model, building a deep
reinforcement learning algorithm framework, initializing a neural
network strategy, designing a layered reward function, and collecting the state information of the unmanned aerial vehicle and the on-load mechanical arm. The strategy neural network outputs a stability augmentation control instruction according to the collected state information, the stability augmentation control instruction and a control instruction output by the
PID controller are subjected to weight superposition with parameters, a fusion control law is obtained, then the fusion control law is input into the unmanned aerial vehicle and on-load mechanical arm model, next state information is solved, the next state information is used for calculating a
reward value of the state, and the
reward value is obtained. And corresponding actions are fed back to the deep
reinforcement learning algorithm framework, a distributed near-end optimization strategy is operated to output a strategy gradient, the
gradient descent of the strategy is calculated, the neural network is updated by using the
gradient descent, and strategy updating is completed. The method can give guidance information in real time, and improves the training efficiency.