Autonomous Vehicle Systems in Intelligent Interconnected Transportation Networks

Abstract

This work explores the problem of personalizing the autonomous driving experience, using Advanced Driving-Assistance Systems (ADAS) that combine reinforcement learning (RL) and federated learning (FL) algorithms. Based on the concept of interconnected vehicles with processing capabilities, in this work, we demonstrate how FL can minimize the time needed to train the RL-based personalization models, taking advantage of the collective knowledge of multiple drivers with similar profiles. In order to demonstrate the efficiency of our proposed approach we perform experiments on a driving simulation environment for a dynamic driving mode selection task. The results show that the proposed solution can better balance drivers' stress and excitement, and optimize the route completion time for different driver profiles. The federated solution reduces the overall time needed for model training and provides a promising solution for driving 1 mode personalization in ADAS.In order to provide a holistic view of the challenges for federated learning in collaborative scenarios, we also discuss the risk of attacks from adversarial users on private and sensitive data that are used in this process.