Shared space multi-modal traffic modeling using LSTM networks with repulsion map and an intention-based multi-loss function

Abstract

Shared space is an innovative, yet controversial concept of traffic regulation that features the coexistence of different types of road users (cars, motorcycles, bicycles, pedestrians, etc.), with the same priority (no traffic lights or signs) and in the same, unsegregated environment. The effect on traffic conditions, as well as on safety, has not yet been systematically documented due to the lack of dedicated traffic models and simulation tools. In this paper, we develop a variation of the Occupancy LSTM network, enhanced with spatial and temporal proximity aspects and a sophisticated, intention-based loss function that models interactions in a shared space environment and predicts the trajectories of vehicle and pedestrian users. The model achieves a significant prediction accuracy, which allows its exploitation for simulating shared space conditions. Moreover, a simulation experiment is designed to evaluate the concept of shared space in terms of efficiency and safety. Results indicate that the interactions that take place in such an environment would lead to a slight increase of travel time for both types of users (compared to free flow conditions) and a low number of possible conflicts.