Acoustic-driven Interior Vehicle Adaptation based on Deep Reinforcement Learning to Improve Driver's ComfortDownload PDF


Apr 13, 2018 (edited Jun 25, 2018)IEEE 2018 ITSC Blind SubmissionReaders: Everyone
  • Abstract: The safety and comfort of drivers have been improved over the decades through new technologies and driver modeling studies that broadened understanding of predicting driver's behaviors. Despite remarkable advances in autonomous systems and interactive systems, there is a significantly lack of approaches considering the passengers and the vehicle as components of a dynamical vibro-acoustical system. The sound in vehicles is not only informative of the state of the vehicle and the environment, but it can also affect driver's performance, attention, and pleasantness of driving. This project aims to investigate the interplay between the perceived sounds of a vehicle and the psychoacoustic annoyance metrics. Our goal is to create an intelligent agent that acts to improve the driver'€™s pleasantness through acoustic-driven learning. To tackle the problem of acting to reduce the annoyance, we present a method based on reinforcement learning based that learns from the environment, i.e., the vehicle interior. Our method changes the state inside the vehicle (closing or opening the window and choosing the cruise speed) to avoid annoying sounds in its interior. The results of this work, all performed using the GTA V simulator, showed that the trained agent learned to take actions to avoid creating annoying sounds.
2 Replies