Prediction of Traffic Participant Behavior in Driving Environment Based on Counterfactual Inference

To travel safely and efficiently in complex traffic scenarios, autonomous vehicles need to be able to accurately predict the intentions and future behaviors of surrounding traffic participants. Nowadays, the behavior prediction of autonomous driving is mostly achieved using machine learning and deep learning models. However, due to the complexity and opacity of these models, it is difficult to meet the transparency requirements of autonomous driving systems in terms of interpretability. Therefore, by changing the input data, that is, using the mask to add or delete the traffic participants in the driving scene and modify the operation of the lane structure, so as to explore the main factors affecting the behavior prediction and further improve the interpretability of the model. To verify the interpretability of counterfactual inference operations for behavioral prediction methods, this paper conducts experiments on a large-scale autonomous driving public dataset nuScenes. Through a detailed analysis of the experimental results, we can conclude that the proposed method is interpretable for the behavioral prediction results.