Bilateral Adaptation of Longitudinal Control of Automated Vehicles and Human Drivers

Abstract

Automated vehicles have great potential to transform our existing transportation systems by improving driving safety, comfort, congestion, and emissions. Despite the tremendous efforts that have been spent on the development of various automated driving technologies, user acceptance of automated driving technologies is still low, which is largely caused by the gaps between automated driving controllers and human preferences. Recent research efforts have been focusing on adapting automated driving behaviors to human demonstrations. However, most existing methods assume that human demonstration is perfect and only focus on mimicking human driving behaviors. In reality, the human demonstration will not be ideal and will include some over-aggressive or over-conservative actions that compromise the safety or efficiency of the trained automated driving controller. In this paper, an Inverse Model Predictive Control (IMPC) based bilateral adaptation method for automated vehicles and human drivers is proposed. The method can adapt automated longitudinal driving behaviors to human preferences based on human interventions during automated driving. Meanwhile, it can also reject improper interventions and send warnings to the human driver such that he/she can realize the irrationality in his/her behaviors. Eventually, the automated driving controller will adapt to the human driver’s preferences and the human driver will get rid of his/her bad driving habits. Human-in-the-loop experiments were conducted using a driving simulator to demonstrate the effectiveness of the proposed approach.