Mixture-of-Behaviors and Levels-of-Expertise in a Bayesian Autonomous Driver Model

Traffic scenario simulations and risk-based design require Digital Human Models (DHMs) of human control strategies. Furthem10re, it is tempting to prototype assistance systems on the basis of a human driver model cloning an expert driver. We present the model architecture for embedding probabilistic models of human driver expertise with sharing of behaviors in different driving maneuvers. These models implement the sensory-motor system of human drivers in a mixtureof-behaviors (MoB) architecture with autonomous and goal-based attention allocation processes. A Bayesian MoB model is able to decompose complex skills (maneuvers) into basic skills (behaviors) and vice versa. The Bayesian-MoB-Model defines a probability distribution over driver-vehicle trajectories so that it has the ability to predict agent's behavior, to abduct hazardous situations, to generate anticipatory plans and control, and to plan counteractive measures by simulating counterfactual behaviors or actions preventing hazardous situations.