Risk Assessment and Management based on Neuro-Fuzzy System for Safe and Flexible Navigation in Unsignalized Intersection

This paper proposes an Unsignalized Intersection Management Control Strategy (UIM-CS) to enable an autonomous vehicle to perform a safe and smooth maneuver, taking into account the curvilinear trajectories of the considered vehicles. This is done while using a metric to assess the risk of the encountered situation through the appropriate use of the Predicted Inter-Distance Profile (PIDP) [1], [2], and its controlled minimum (mPIDP). The proposed control is based on an adaptive PD controller where the parameters are learned by using an Adaptive Network based Fuzzy Inference System (ANFIS). The variables that allow the assessment of the dangerousness based on PIDP are carefully defined to allow the genericity of the approach to all types of insertions, especially the unsignalized one (e.g., roundabout or highway insertion) where the Autonomous Vehicle (called Ego-Vehicles (EVs) in what follows) has to make a choice on its behavior (acceleration/deceleration). The proposed approach for the creation of the dataset allowing the learning of the adaptive PD controller parameters, that directly influence the responsiveness of the EV while taking into account its actual capacity and constraints, is also presented. To demonstrate the reliability and safety of the overall proposed control architecture, several simulations are performed.