The increasing adoption of end-to-end learning for autonomous driving introduces increased model complexity and opacity, raising the risk of learning undesired or erroneous behavior. This work integrates unsupervised dictionary learning as a post hoc interpretability module within state-of-the-art driving models to decompose driving behavior into semantically meaningful concepts while demonstrating their causal influence on the model's driving decisions.
We propose a stepwise framework for extracting and interpreting meaningful concepts from the end-to-end model and connecting them to the multifaceted model outputs, thereby revealing the underlying decision-making logic for the prediction of future trajectories. Furthermore, targeted interventions at the concept level allow us to manipulate and correct driving decisions, resulting in measurable improvements in overall driving performance.
Thus, we demonstrate how interpretability can effectively reduce model opacity, uncover erroneous behavior, and enable targeted mitigation, ultimately boosting model performance.
Blogger's Review: This study presents an innovative approach to interpretability by integrating unsupervised learning with end-to-end autonomous driving models, effectively elucidating the logic behind model decisions. This not only enhances model transparency but also provides stronger safety assurances for future autonomous driving systems.