Treffer: Disseny, configuració i validació d'algorismes de Machine Learning aplicats la gestió automatitzada d'encaminament per a vehicles autònoms amb descàrregues de continguts en un entorn controlat

This Work addresses the design, implementation and validation of an autonomous vehicle routing system designed to be used in controlled urban environments. The project places technologies such as cloud computing, edge computing and artificial intelligence, especially deep reinforcement learning, as the foundations of the proposed solution. Initially, there is a design stage where the characteristics required by the proposed solution are specified, and the scenario where the field tests have been carried out is also specified. Next, the various blocks defined using Python programming and various devices of the NVIDIA Jetson family have been implemented. The incorporation of artificial intelligence has been left to take place once the routing system has been validated. At this point, an agent and an algorithm have been developed to incorporate the entire system developed as an environment accessible by the agent. Finally, several tests have been carried out that have consisted of training the agent created using the routing system as an environment with the aim of validating the operation of the entire set of functional blocks, carrying out tests based on both simulations and real tests by deploying the devices used in streetlights and using an autonomous vehicle prototype (also developed in this work). The tests carried out have made it possible to corroborate that the system works correctly, following the specifications defined at the beginning of the work. In addition, the system has shown that developing autonomous vehicle guidance mechanisms is a viable solution, achieving that an artificial intelligence model decides how the vehicles should move around the scenario in real-time.