A new study argues that a carefully designed greedy heuristic can outcompete sophisticated optimization algorithms when routing vehicles in real time.
The paper, posted to arXiv on July 8, 2026, introduces the Efficiency heuristic — a reward-density scoring rule for assigning tasks to a fleet of vehicles as new jobs arrive mid-operation. Researchers tested it against four classical construction heuristics and three metaheuristics: Adaptive Large Neighbourhood Search, a Genetic Algorithm, and Simulated Annealing. Across drone task allocation and urban taxi dispatch scenarios, at multiple fleet sizes and task volumes, the Efficiency heuristic matched the best metaheuristic on total reward collected. The catch: it did so while requiring two to three orders of magnitude less planning time.
That gap matters because real-time dispatch systems live and die by latency. A solver that finds a marginally better route but takes 500 milliseconds to do it is useless if new tasks arrive every 50 milliseconds. The finding echoes a recurring pattern in operations research — that the theoretical optimality ceiling of a method and its practical value in online settings are very different things.
The result is not entirely surprising to anyone who has watched the robotics taxi and drone-delivery industries wrestle with exactly this tradeoff, but it is a useful empirical reminder that reaching for a metaheuristic first is not always the right engineering instinct.