Risk Representation, Perception, and Propensity in an Integrated Human Lane-Change Decision Model
Longsheng Jiang, Dong Chen, Zhaojian Li, Yue Wang
Abstract
Lane-change decision models with enhanced human-likeness are increasingly important as they are integral in traffic simulations for training autonomous driving algorithms. This work proposes a computational model of driver lane-change decision-making by integrating relevant human features in perception, reasoning, emotion, and decision (PRED). The PRED model describes how drivers make lane-change decisions under collision risk. Here risk is represented by probabilities and outcomes of the possible consequences. The PRED model formulates drivers’ risk perception and risk propensity in its modules: the perception module is modeled with Bayesian inference; the reasoning module is modeled with Newtonian simulation; the emotion and decision module is modeled with the extended regret theory. The PRED model was fitted and tested with an empirical dataset from a naturalistic driving database. The prediction performance of the PRED model ranks higher than the selected benchmarks and is close to the state-of-the-art machine learning models. Moreover, the explicit modeling of risk propensity sheds light on an important question in transportation: what causes human drivers’ risk-taking behaviors? The results support the rationale that downplaying crash consequences is the main contributor.