A Decision Making Model for Overtaking Maneuvers in Autonomous Driving Through Driver Behavior Analysis

Abstract

There are a myriad of stationary vehicles in urban driving environments. It is necessary to decide whether or not to overtake a stationary vehicle in autonomous driving. To solve this decision-related problem, previous studies have depended on the parking classification of stationary vehicles. However, there is a limitation in this approach- it is applicable to the limited situations where the stationary vehicle can obviously be inferred as a parked vehicle. This does not reflect the characteristics of the human drivers’ decisions, in which not only the preceding vehicle but also the context of the surrounding situation is considered.

To overcome the limitation, this work proposes a DNN-based decision making model that learns a situation-awareness decision making strategy from an expert driver. The paper describes two processes for designing the model. The first process is analyzing and selecting the primary factors that a driver considers when deciding on overtaking maneuvers. The second process is constructing a DNN model of which input features are the dominant factors selected in the previous process. Through these processes, the model was able to make situation-awareness decisions similar to that of the expert driver.

The decision making performance of the proposed model was evaluated in various traffic scenarios. In the scenarios where the preceding vehicle was not inferred as a parked vehicle, the proposed model was able to decide to overtake the vehicle. It showed the model’s ability to consider the surrounding traffic situations. By quantitatively comparing the decision results of the model and those of the expert driver, we could conclude that the model makes situation-awareness decisions in uncertain situations in the similar manner to the driver.