Deep Neural Network-Based UAS Transport

The paper develops a deep neural network- (DNN) based mass transport approach to cover a distributed target in a decentralized manner by Uncrewed Aerial Systems (UAS). This is a new decentralized UAS transport approach with time-varying communication weights that can be achieved by solving the following three sub-problems: (i) determining the DNN structure, (ii) obtaining communication weights, and (iii) ensuring stability and convergence guarantee. By proposing a novel algorithmic approach, the DNN is structured based on the UAS initial formation with an arbitrary distribution in the motion space. To specify communication weights for a team of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$N$</tex> multi-copters, we use the DNN to obtain the initial communication weights, based on the agents' initial positions, abstractly represent the distributed target by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$N$</tex> points, considered as the final positions of all agents, and obtain the final communication weights. The third sub-problem is to prove the stability and convergence of the UAS transport.