A Deep Learning Framework For Modeling Pedestrian – Vehicle Interactions In Shared Space

Shared Space is a novel, laneless traffic design paradigm, where different types of users (motorized vehicles, bicycles, pedestrians, etc.) can move freely and there are no priority rules. Thus, a significant number of complex and safety critical interactions between the users. In this paper, a Deep Learning, LSTM-based model is proposed for estimating the trajectory of heterogeneous users, enhanced with an efficient feature extraction strategy, which is trained using naturalistic trajectory data. Results indicate the suitability and the potential of the proposed approach for modeling shared space interactions, as the accuracy metrics are lower than most state-of-the-art models. In the future, the model will also be used to develop a simulator for shared space environments.