Design and development of a Vehicular Digital Twin based on Reinforcement Learning

Abstract

This project evaluates the feasibility of employing an interurban vehicular Digital Twin, replicating a section of the C-32 highway in Catalonia, for training a reinforcement learning system. A vehicular Digital Twin, which utilizes Aimsun Next software tool to simulate the digital replica of the real world and employs strategies to alleviate congestion, has been adapted to function with reinforcement learning systems. Subsequently, a range of reinforcement learning systems have been developed, featuring systems with varying complexity levels, with the usage of different observation spaces, and action spaces together with the test of key performance indicators for reward computation. Furthermore, the project includes an analysis of the feasibility and the performance of the reinforcement learning systems in decision making tasks.