Decoding pedestrian and automated vehicle interactions using immersive virtual reality and interpretable deep learning

To ensure pedestrian friendly streets in the era of automated vehicles, reassessment current policies, practices, design, rules and regulations urban areas is importance. This study investigates crossing behaviour, as an important element dynamics that expected to be affected by presence vehicles. For this purpose, interpretable machine learning framework proposed explore factors affecting pedestrians' wait time before mid-block crosswalks collect rich behavioural data, we developed a dynamic immersive virtual reality experiment, with 180 participants from heterogeneous population 4 different locations Greater Toronto Area (GTA). Pedestrian behaviour then analyzed using data-driven Cox Proportional Hazards (CPH) model, which linear combination covariates replaced flexible non-linear deep neural network. The model achieved 5% improvement goodness fit, but more importantly, enabled us incorporate richer set covariates. A game theoretic based interpretability method used understand contribution pedestrians crossing. Results show vehicles on roads, wider lane widths, high density limited sight distance, lack walking habits are main contributing longer times. Our suggested that, move towards pedestrian-friendly areas, national level educational programs for children, enhanced safety measures seniors, promotion active modes transportation, revised traffic should considered.