Modeling and Analysis of Human Drivers' Compliance Behavior to Maneuver Recommendations in Form of Soft Inputs
Studies have shown that the benefits of fully auto-mated traffic, namely increased safety and efficiency, are being negated in mixed traffic scenarios already by a low percentage of human drivers. Besides human driving and perception errors, this effect is caused by the lack or coarseness of communication between the automated intersection control system and the human drivers, typically only through classical traffic lights. In this work, we propose to extend the interaction capabilities between human-driven vehicles (HDVs) and an automated intersection with warnings and maneuver recommendations, such as lane-change or velocity recommendations (comp. green light optimal speed advisory (GLOSA)) or safety warnings. These so-called soft inputs enable HDVs to travel more efficiently and safely by exploiting 5G communication, collective perception, and state-of-the-art intersection control concepts. The human compliance behavior to such inputs is modeled, calibrated, and simulated based on expert knowledge and naturalistic driving data. A suitable maneuver recommendation formulation for a conflicting unprotected left-turn scenario is outlined and tested in simulation studies in a typical urban intersection scenario. Analyzing relevant key performance indicators shows the high achievable performance and its trade-off characteristics with respect to HDV penetration rate and compliance behavior.