Effective management of disaster-induced traffic disruptions is critical to reducing emissions, economic losses, and emergency-response delays in increasingly congested urban environments. This study leverages the combined power of UAVs and Grammar-Guided Genetic Programming (G3P) to evolve a rerouting cost function for disaster scenarios and benchmarks its performance against existing UAV-based rerouting algorithms originally designed for full (100%) smart-vehicle adoption. Using SUMO simulations of Dublin City Centre under five distinct disaster events and five levels of UAV-derived information, we evaluate RR G3P at 100%, 70%, and 50% smart-vehicle adoption rates and test its transferability to adjacent road segments. At 100% adoption, RR G3P outperformed baselines in 96% of cases (24/25), compared to 44% (11/25) at 70% and 24% (6/25) at 50%, demonstrating robust performance under high compliance but reduced adaptability with fewer connected vehicles. Transferability experiments yielded mixed results: while RR TransferredG3P occasionally outperformed all baselines, algorithms such as RR Length showed consistent, context-specific strengths across varied UAV coverage levels. We observed a maximum improvement of 44.23% in Average Arrival Time compared to the no-rerouting scenario, and 14.99% compared to the non-disaster scenario at full smart-vehicle adoption, however once the adoption level drops to 70%, RR G3P performance drops by 4.53% (Disaster 4, Information Level 4) and 10.45% once the adoption further drops to 50%. These findings indicate that, although RR G3P is highly effective in its tailored training environment, further refinement—particularly through analysis of segment-specific dynamics—is necessary to ensure broad applicability in disaster-impacted urban networks.
Wlodarczyk et al. (Sun,) studied this question.