D-ACC: Dynamic Adaptive Cruise Control for Highways with On-Ramps Based on Deep Q-Learning
share
An Adaptive Cruise Control (ACC) system allows vehicles to maintain a desired headway distance to a preceding vehicle automatically. ACC is increasingly adopted by recently available commercial vehicles. Recent research demonstrates that effective use of ACC can improve the traffic flow by adjusting the headway distance in response to dynamically changing traffic conditions. In this paper, we demonstrate that state-of-the-art real-time ACC systems may perform poorly in highway segments with on-ramps because their simple model based only on the traffic conditions of the main road does not take into account the dynamics of merging traffic in determining the optimal headway distance. We propose D-ACC, a dynamic adaptive cruise control system based on deep reinforcement learning that effectively adapts the headway distance according to dynamically changing traffic conditions of both the main road and merging lane to optimize traffic flow. Extensive simulations are performed with a combination of a traffic simulator SUMO and vehicle-to-everything communication (V2X) network simulator Veins under various traffic scenarios. We demonstrate that D-ACC improves the traffic flow by up to 70 in a general highway segment with an on-ramp.
READ FULL TEXT