Högskolan i Skövde

Progress in artificial intelligence (AI), onboard computation capabilities, and the integration of advanced sensors in cars have facilitated the development of Advanced Driver Assistance Systems (ADAS). These systems aim to continuously minimize human driving errors. {An example application of an ADAS could be to support a human driver by informing if an intended driving maneuver is safe to pursue given the current state of the driving environment. One of the components enabling such an ADAS is recognizing the driver's intentions. Driver intention recognition (DIR) concerns the identification of what driving maneuver a driver aspires to perform in the near future, commonly spanning a few seconds. A challenging aspect of integrating such a system into a car is the ability of the ADAS to handle unseen scenarios. Deploying any AI-based system in an environment where mistakes can cause harm to human beings is considered a high-risk AI system. Upcoming AI regulations require a car manufacturer to motivate the design, performance-complexity trade-off, and the understanding of potential blind spots of a high-risk AI system.} Therefore, this licentiate thesis focuses on AI-based DIR systems and presents an overview of the current state of the DIR research field. Additionally, experimental results are included that demonstrate the process of empirically motivating and evaluating the design of deep neural networks for DIR. To avoid the reliance on sequential Monte Carlo sampling techniques to produce an uncertainty estimation, we evaluated a surrogate model to reproduce uncertainty estimations learned from probabilistic deep-learning models. Lastly, to contextualize the results within the broader scope of safely integrating future high-risk AI-based systems into a car, we propose a foundational conceptual framework.

Deep learning (DL) methods have advanced rapidly and are commonly applied in high-risk, resource-constrained environments such as advanced driver assistance systems (ADAS), where misclassifications can have serious consequences. With upcoming artificial intelligence (AI) legislation, it is essential to extensively evaluate and minimize the undesirable behavior of DL-based systems in such settings. An example is an ADAS that continuously evaluates whether a driver’s intended maneuvers are safe to execute given the current traffic context. Driver intention recognition (DIR), which predicts the maneuver a driver intends to perform in the near future, is a central DL-based component of such systems. Since deep neural networks (DNNs) do not inherently provide uncertainty estimates for their predictions, probabilistic deep learning (PDL) methods can be applied to improve the identification of scenarios where model outputs may be unreliable. In this thesis, we first review the current state of DIR research, focusing on the recent shift toward DL methods. We then examine how both established and novel PDL methods influence DIR performance. We evaluate the uncertainty estimations by analyzing their ability to distinguish between correct and incorrect predictions and by measuring their effectiveness in out-of-distribution (OOD) detection. Furthermore, we employ neural architecture search with multiple objectives and search strategies to explore how architectural complexity impacts DIR and OOD detection performance. Finally, we conduct a comparative experiment to evaluate human performance against that of DL-based models in video-based recognition of road user intentions.