Optimal Estimation Using Deep Neural Networks Applied to Navigation and Motion Control

Abstract

Abstract The critical analysis is given concerning the current state of using deep neural networks with convolutional and recurrent layers, a recurrent network of Long Short-Term Memory, Gated Recurrent Units for estimation tasks in relation to navigation and motion control. A comparison of neural network and traditional methods is given for understanding and explaining their functioning. The differences, advantages and disadvantages of deep neural networks in relation to solving estimation problems are revealed. The possibility of machine training with reinforcement is analyzed for estimation tasks in navigation and motion control in real time. The prospects of using neural networks in the processing of navigation data, as well as for the tasks of adaptive estimation and trajectory tracking, are formulated.