Building a Situation Awareness Framework for Communicating Among Road Users in Mixed Automated and Manual Traffic Environments

Abstract

On the promise of reducing human error and increasing road safety, the number of autonomous vehicles (AVs) in the industry and on the roads is steadily rising. However, this technology faces challenges in various scenarios, such as interactions with AV drivers or in mixed traffic environments where AVs share the road with conventional vehicles (CV), pedestrians, and bicyclists. In these situations, situation awareness (i.e., the ability to perceive, comprehend, and predict the situation on the road) is crucial to ensure road safety for all road users. To address this challenge, this dissertation aims to design and evaluate explanations based on the theory of mind to improve human-machine performance. The central hypothesis is that explanations during human-machine interaction can provide the necessary information to resume situation awareness, improving the joint performance of the human-machine team. To achieve this objective, three fundamental questions were investigated: (1) how should the AV assess the driver's situation awareness to understand the need for explanations, (2) how should the AV driver understand the state of mind of the AV through explanations, and (3) how should the AV share information with CV drivers through explanations. To tackle these questions, first, a machine learning model was developed to predict the situation awareness of AV drivers in real-time using behavioral, physiological, and self-reported data. A LightGBM (Light Gradient Boosting Machine) model trained on the most critical predictors identified by SHAP (SHapley Additive exPlanations) achieved promising performance.Next, an explanation framework was proposed based on the situation awareness model and explainable AI. The framework was tested in an online environment by evaluating participants' situational trust, cognitive workload, and explanation satisfaction. The results showed that properly designed explanations based on the proposed framework assisted drivers in unexpected situations, increased their trust in AVs, and improved their situation awareness. Finally, external and internal HMI concepts were proposed to explore the interaction between AVs and CVs in challenging situations and improve situation awareness among CV drivers. The concepts were tested in a virtual reality environment using self-reported and physiological measurements. The findings revealed that explanations were able to increase participants' situation awareness and trust in AVs, with the internal HMI perceived as the most effective. Overall, this research aims to contribute new fundamental knowledge about how to build situation awareness to improve human-machine performance by designing and evaluating human-centered explanations, particularly in conditionally automated driving. The research also seeks to enhance the relationship between humans and technology in automated driving and other fields, such as manufacturing and medical industries.