Agent-Based Modelling of Dynamics of Interacting Unmanned Ground Vehicles Using FLAME GPU

An approach to modelling traffic flows based on an intelligent analysis of dynamics of unmanned ground vehicles (UGVs) is proposed. Within this approach, a novel decision-making system for agent-UGV manoeuvring is designed. The proposed system uses clustering methods for estimating traffic congestion density on alternative routes of the digital road network (DRN) to detour congested areas and minimize risks of traffic accidents. For the first time, the spatial dynamics of interacting UGVs is modelled for complicated DRNs, which include multiple intersecting ring, straight, and diagonal segments that form various alternative routes for road users. The developed model is implemented within the supercomputer agent-based modelling framework FLAME GPU. An algorithm for the behaviour of UGVs during the formation of traffic jams is developed, which, in particular, provides an effective splitting of traffic flows by increasing the probability of choosing less congested alternative routes for UGVs. As a result of numerical experiments, the following important regularity was revealed: an increase in the proportion of UGVs in the DRN using intelligent manoeuvring based on cluster analysis of traffic flows with a controlled radius of an agent personal space provides a decrease in the total density of traffic congestion and contributes to a decrease in the number of accidents.