Urban waste collection remains one of the most resource-intensive services that a city must provide, and in most municipalities, it is still carried out along fixed routes that take no account of how full the bins actually are. Trucks visit containers...
Urban waste collection remains one of the most resource-intensive services that a city must provide, and in most municipalities, it is still carried out along fixed routes that take no account of how full the bins actually are. Trucks visit containers that are nearly empty while bins in busier districts overflow before the next scheduled round, wasting fuel, labor, and vehicle hours and leaving residents dissatisfied. The growing availability of low-cost fill-level sensors has made it possible to know the state of every bin in close to real time, yet that information is of little value unless the routing system is able to act on it. This thesis addresses that gap by designing and implementing a dynamic vehicle routing system that continuously adapts collection routes to the live condition of an instrumented bin network.
The system was built as a complete, working application for the Haeundae district of Busan, comprising one hundred bins, four collection trucks operating from four separate depots, and four disposal facilities. At its center is a routing engine that treats the task as a multi-depot capacitated vehicle routing problem and re-plans whenever a bin crosses an eighty-percent fill threshold. To determine which routing strategy is best suited to this setting, ten algorithms were implemented and compared on identical data under a fixed random seed, ranging from simple construction heuristics such as Nearest Neighbour and the Clarke-Wright savings method, through classical metaheuristics including Simulated Annealing, Tabu Search, Genetic Algorithms, Ant Colony Optimization and Large Neighbourhood Search, to the constraint-programming solver provided by Google OR-Tools. Each was evaluated on total travel distance, fuel consumption, and computation time, and the benefit of dynamic re-routing was measured separately against a static plan.
The results show a clear and consistent ordering. On the shared thirty-bin instance the OR-Tools solver produced the shortest routes at 24.66 kilometres, roughly thirty-five percent less than the 37.71 kilometres travelled by the Nearest Neighbour baseline, with a corresponding reduction in fuel from 12.63 to 8.24 litres. Metaheuristics occupied the middle of the ranking, while the pure construction heuristics performed worse, confirming that the additional search effort invested by the stronger methods translates into real savings on this problem. The dynamic experiments demonstrated that re-routing in response to newly full bins allowed the fleet to service fifty percent more containers than a static plan without a proportional increase in distance. Taken together, the work shows that combining an instrumented bin network with a capable routing solver and a dynamic re-planning policy yields a measurable and reproducible improvement in collection efficiency, and that the choice of routing algorithm is itself a decision worth making with care.