Developing deep LSTM model for real-time path planning in unknown environments

This paper proposed a real-time path planning approach based on deep LSTM model, in which the model was constructed by Long Short-Term Memory (LSTM). By collecting various scenarios and paths generated by the existing static path planning algorithm, we form the dataset including the UAV's optimal behaviors and the current observed threats at each time step, which is utilized for training the deep LSTM model. Thus, through learning the training dataset, the model can make optimal decisions in real-time according to the threat around at the current time. Experimental results show that the proposed approach has the ability to fulfill the real-time path planning requirements, including environment unknown, navigation to the target and collision-free. Moreover, our real-time path planning approach can improve the global optimization quality of the path.