Comparison of algorithms for multi-agent pathfinding in crowded environment

Abstract

The objective of this paper is to recommend the algorithms for planning the best paths for the simultaneously moving agents operated in the large crowded environment. The proposed approach consists of two parts. Firstly, a navigation mesh for passable regions in rectangular 2D environment is created using Quad-trees algorithm. A graph is created by connecting centers of regions, where weights in the graph are Euclidean distances between centers. In the second part, a path is found for each agent currently present in environment using Dijkstra or A* algorithm. To plan good path in each passable region actual density value is stored. Density information is further mapped on graph edges along with distance value. Agents reevaluate their paths accordingly to re-planning strategy. Three strategies are considered: periodical re-planning, periodical with initial re-planning and event-driven re-planning proposed by the authors. The six combinations of algorithms have been implemented and tested. Simulation experiments made using the created experimentation system showed that the approach with the own event-driven re-planning seems to be promising.