A Deep Reinforcement Learning Approach for Efficient, Safe and Comfortable Driving

Abstract

Sensing, computing, and communication advancements allow vehicles to generate and collect massive amounts of data on their state and surroundings. Such richness of information fosters data-driven decision-making model development that considers the vehicle’s environmental context. We propose a data-centric application of Adaptive Cruise Control employing Deep Reinforcement Learning (DRL). Our DRL approach considers multiple objectives, including safety, passengers’ comfort, and efficient road capacity usage. We compare the proposed framework’s performance to traditional ACC approaches by incorporating such schemes into the CoMoVe framework, which realistically models communication, traffic, and vehicle dynamics. Our solution offers excellent performance concerning stability, comfort, and efficient traffic flow in diverse real-world driving conditions. Notably, our DRL scheme can meet the desired values of road usage efficiency most of the time during the lead vehicle’s speed-variation phases, with less than 40% surpassing the desirable headway. In contrast, its alternatives increase headway during such transient phases, exceeding the desired range 85% of the time, thus degrading performance by over 300% and potentially contributing to traffic instability. Furthermore, our results emphasize the importance of vehicle connectivity in collecting more data to enhance the ACC’s performance.

Keywords

• connected vehicles
• vehicle dynamics
• adaptive cruise control
• traffic simulation
• machine learning