Towards Trustworthy Automated Driving through Qualitative Scene Understanding and Explanations

Artificial Intelligence (AI) methods nowadays are at the center of automated driving and connected mobility, including perception and scene understanding [1, 2, 3]. However, passing control to an AI-based system and trusting its decisions requires the ability to request explanations for these decisions [4]. Societal acceptance of automated driving significantly depends on these AI models' trustworthiness, transparency, and reliability [5]. Still, this is an open challenge, as many of the state-of-the-art machine learning (ML) models are opaque and not inherently explainable by themselves [6]. In recent years, several explainable AI methods with a focus on automated driving have been proposed. Following [6], they fall into three main categories: a) Vision-based explainable AI related to highlighting the area of an image that influences a perception model towards a certain output [4]; b) Feature-based importance scores quantify the influence of each input feature on the model output; and c) Textual-based explainable AI that aims to formulate explanations as intelligible arguments using natural language processing [7]. Unfortunately, automated support for multisensor and video-based scene explanation is still restricted to quantitative analysis, e.g., saliency heatmaps [4]. In this work, we exploit qualitative methods for scene understanding by using Qualitative Explainable Graphs (QXG) and, based on this representation, we propose a method for action explanation through simple classification models.