Route Orchestrator

Multi-agent path finding (MAPF) computes collision-free routes for many agents on a shared network — a core problem for autonomous logistics and intelligent transportation. Applying it to real cities requires both realistic map data and validation in a microscopic traffic simulator.

Route Orchestrator plans on OpenStreetMap networks processed with OSMnx, using Conflict-Based Search (CBS): a custom Time-Aware A* finds each agent's route in space–time (including wait actions), while the high-level Conflict Tree resolves vertex and edge conflicts. Local agent movement follows Cellular-Automata-inspired rules (acceleration, braking, random deceleration, segment speed limits from OSM).

The main contribution is a hybrid OSM→SUMO route conversion that maps routes planned on the simplified OSMnx graph onto SUMO's more detailed network — combining direct node matching, controlled use of sumolib and ordering by OSM edge origIDs to resolve ID and segmentation mismatches. Visual tools cover agent animation, step-by-step conflict resolution, space–time diagrams and performance scaling.

Experiments on the road network around 23 Nisan Street in Bursa (10 random start/goal sets per agent count, node-expansion limits of 500 and 3,000) solved all instances with up to five agents (about 6.7 s computation); success fell as agent counts grew (51/60 for 10–15 agents and 19/100 for 21–30 agents with the higher limit), motivating more scalable CBS variants. Presented at IEEE ASYU 2025, Bursa.