Computational human driving behavior reference models for ADS evaluation based on (en)active inference
Title: Computational human driving behavior reference models for ADS evaluation based on (en)active inference
Abstract: Computational models of human road user behavior plays a key role in the development and evaluation of automated driving systems (ADS), both for representing other road users in simulation and for establishing human performance benchmarks for ADS. This talk will focus on the latter application, introducing the general concept of Driving Behavior Reference Models (DBRM) for ADS evaluation and presenting recent work at Waymo and our academic partners in this area. In particular, the talk will discuss how human road user behavior can be generally framed based on principles of biological intelligence developed in enactive cognitive science and the free energy principle, and implemented computationally in simulated agents as active inference process models.