Deep Learning for Autonomous Vehicle Route Optimization in Rural Areas

Abstract

We attempt to describe a deep neural network that is used to optimize the route selection of the AUTMOTH. The proposed DRLNDT network optimizes the route by making use of a decision transformer. This decision transformer takes a route graph as an input and outputs the future action sequence tiers. This mechanism works better than the transformer mechanism for route selection due to those action sequence tier outputs and results in a simpler task for the DRL task learner. Furthermore, we use a deep reinforcement learning mechanism to train the DRL task learner in-order that the actions of the next time step can be predicted in an efficient manner. We train our network using a graph based environment where we divide the complete environment into grids in-order to save the training computational time.