On the Applicability of Reinforced Learning in the Route Selection Task of an Unmanned Vehicle
Mikhail G. Gorodnichev · 2023
This paper presents the process of describing and investigating methods for solving the optimal route search problem, choosing the means for developing a software module, including consideration of the required tools, description of the project structure, and features of the algorithm used. An introduction to the field of optimal route finding is given through the basic terms of graph theory, which is a leading framework for the field of reinforcement learning. The process of developing a learning environment is reviewed, starting with the selection of the environment and an explanation of its possible properties. Then the research of selected multi-agent learning algorithms is described in order to make a final comparison according to the most important criteria. Neural network architectures, hyperparameter learning tables and quality graphs of the learning models are plotted.