DRIVE: Dependable Robust Interpretable Visionary Ensemble Framework in Autonomous Driving

TL;DR

The paper tackles the instability and opacity of end-to-end autonomous driving models by introducing DRIVE, a Dependable Robust Interpretable Visionary Ensemble Framework that enforces four properties—Consistent Interpretability, Stable Interpretability, Consistent Output, and Stable Output—through a multi-objective optimization that integrates surrogate top-$k$ losses and a PGD-based training regime. Building on the DCG concept bottleneck, DRIVE preserves pre-trained parameters while achieving more stable explanations and predictions under perturbations, leveraging a Longformer-based temporal encoder and multi-modal encoders (image and text) for scenario alignment. Empirical results on Comma2k19 show that DRIVE outperforms DCG in both predictive accuracy (MAE) and interpretability stability (Top-$k$ overlap) across perturbation types, indicating improved robustness and trustworthiness. The findings lay groundwork for safer, regulatory-aligned autonomous driving systems and point to future work expanding DRIVE to more diverse environments and real-world deployment.

Abstract

Recent advancements in autonomous driving have seen a paradigm shift towards end-to-end learning paradigms, which map sensory inputs directly to driving actions, thereby enhancing the robustness and adaptability of autonomous vehicles. However, these models often sacrifice interpretability, posing significant challenges to trust, safety, and regulatory compliance. To address these issues, we introduce DRIVE -- Dependable Robust Interpretable Visionary Ensemble Framework in Autonomous Driving, a comprehensive framework designed to improve the dependability and stability of explanations in end-to-end unsupervised autonomous driving models. Our work specifically targets the inherent instability problems observed in the Driving through the Concept Gridlock (DCG) model, which undermine the trustworthiness of its explanations and decision-making processes. We define four key attributes of DRIVE: consistent interpretability, stable interpretability, consistent output, and stable output. These attributes collectively ensure that explanations remain reliable and robust across different scenarios and perturbations. Through extensive empirical evaluations, we demonstrate the effectiveness of our framework in enhancing the stability and dependability of explanations, thereby addressing the limitations of current models. Our contributions include an in-depth analysis of the dependability issues within the DCG model, a rigorous definition of DRIVE with its fundamental properties, a framework to implement DRIVE, and novel metrics for evaluating the dependability of concept-based explainable autonomous driving models. These advancements lay the groundwork for the development of more reliable and trusted autonomous driving systems, paving the way for their broader acceptance and deployment in real-world applications.

Figures (2)

• Figure 1: Top right and bottom left figures show that the interpretable and predicted outputs of DCG are sensitive to perturbations, and on the contrary, our optimization framework DRIVE (bottom right) can cope with this very well.
• Figure 2: Overall pipeline of DRIVE. The input is processed by a feature extractor and a temporal encoder, followed by a concept bottleneck with scenario encoding. The DRIVE model incorporates a multi-objective optimization process, balancing consistent interpretability (Ci), stable interpretability (Si), consistent output (Co), and stable output (So) through auxiliary loss functions. The model is trained using PGD to enhance robustness against perturbations while maintaining interpretability and predictive consistency.