Social Cascade FNN: An Interpretable Learning-Based Decision-Making Framework for Autonomous Driving in Lane Changing Scenarios

Lane changing behavior causes a considerable proportion of traffic accidents. Effective decision-making strategies for autonomous vehicles are promising to enhance traffic safety in lane changing scenarios. Naturalistic driving datasets driven deep learning has emerged as a competitive approach to making lane changing decisions, which is capable to consider social interactions, however, the lack of interpretability hinders its application in safety-critical autonomous driving. To address this issue, we proposed a learning-based lane changing decision-making framework that extracts rules from naturalistic driving datasets. The proposed method employed a cascade Fuzzy Neural Network (FNN) to learn from sequential data, coupled with a social pooling layer that extracts interactions among vehicles. By integrating both temporal and spatial information, this framework enhances the learning ability of the system while preserving the interpretability of FNN. Our method out-performs state-of-the-art approaches on two publicly available datasets, demonstrating its effectiveness in lane changes. The method can also accurately make decisions in diverse driving scenarios and provide decision rules.