XAI for Transparent Autonomous Vehicles: A New Approach to Understanding Decision-Making in Self-Driving Cars

• Published in: International eConference on Computer and Knowledge Engineering (Online) pp. 194 - 199
• Main Authors: Hosseini Azad, Maryam Sadat, Hamidi Imani, Amir Abbas, Shokouhi, Shahriar Baradaran
• Format: Conference Proceeding
• Language: English
• Published: IEEE 19.11.2024
• Subjects: Analytical models; Attention Mechanism; Autonomous automobiles; Autonomous Vehicles; Computational modeling; Decision making; Deep learning; Explainable Deep Driving; Knowledge engineering; Natural Language Explanation; Roads; XAI
• ISSN: 2643-279X
• DOI: 10.1109/ICCKE65377.2024.10874778

While numerous advancements have been achieved in deep learning-based self-driving systems, the lack of transparency, interpretability, and user acceptance remains a significant challenge. Researchers argue that without the ability to explain decision-making behavior, deep learning models cannot be practically implemented in various real-world scenarios. This is vital in decision-making networks, since inaccurate results may cause dangerous road incidents. To address this problem, we propose an innovative approach that integrates the Convolutional Block Attention Module with Deep Connected Attention (DCA-CBAM-ResNet50) with a state-of-the-art decision-maker and textual explainer model. This integration results in a more precise and comprehensive explainable system. To quantitatively analyze the performance of our model, we used the standard F1 score and applied it to the Berkeley DeepDrive-object-induced action (BDD-OIA) dataset. Our proposed technique outperforms the current SOTA model and demonstrates significant improvement in explainability. This research advances future explainable autonomous vehicles and contributes to creating more transparent and trustworthy self-driving systems.