Automatic landing on a moving platform using deep neural network estimation

This paper describes a control architecture designed to enable autonomous landing on a moving vehicle. This architecture can be separated into two parts: prediction and trajectory generation. For the prediction method, we use a Deep Neural Network (DNN), which can estimate multiple future states of the landing target by using past measurement data. For trajectory generation, we employ differential dynamic programming (DDP), which involves the interior penalty function with time-varying weight in the cost function. As a result, we can predict the state for a future time by using the multi-point prediction technique using DNN and generate the trajectory with time-varying weight using DPP. We provide the architecture of DNN and DDP with implementation techniques and sample results.