Traffic simulation with human in the loop: roundabout scenario in a driving simulator

Traffic simulators are powerful tools for the simulation of vehicle interactions in many traffic conditions, but they employ approximated driver models, so the reliability of the acquired information is limited. To overcome such limitations, this paper presents a co-simulation between a widely diffused open source traffic simulator and a high end driving simulator. The coupling of the two simulators is used to investigate the effect of an actual human being driving in a simulated mixed traffic situation, where both traditional and connected and automated vehicles (CAVs) are involved. The selected reference scenario is a three-legged single lane roundabout. The behaviour of the autonomous vehicles in the simulated environment is controlled through a previously trained reinforcement learning policy. The objective is to minimize the time needed to go through the proposed roundabout. Preliminary tests are realized considering a panel of drivers on the driving simulator and with different percentages of autonomous vehicles in the simulation (20% and 80%). Results seem to indicate that the behaviour of CAVs can be easily accepted by human drivers. These outcomes deserve a further extensive statistical investigation for a final assessment.