A faster heuristic for the Traveling Salesman Problem with Drone

Pedro H. D. B. Hokama,Carla N. Lintzmayer,Mário C. San Felice

2024-05-29

Abstract:Given a set of customers, the Flying Sidekick Traveling Salesman Problem (FSTSP) consists of using one truck and one drone to perform deliveries to them. The drone is limited to delivering to one customer at a time, after which it returns to the truck, from where it can be launched again. The goal is to minimize the time required to service all customers and return both vehicles to the depot. In the literature, we can find heuristics for this problem that follow the order-first split-second approach: find a Hamiltonian cycle h with all customers, and then remove some customers to be handled by the drone while deciding from where the drone will be launched and where it will be retrieved. Indeed, they optimally solve the h-FSTSP, which is a variation that consists of solving the FSTSP while respecting a given initial cycle h. We present the Lazy Drone Property, which guarantees that only some combinations of nodes for launch and retrieval of the drone need to be considered by algorithms for the h-FSTSP. We also present an algorithm that uses the property, and we show experimental results which corroborate its effectiveness in decreasing the running time of such algorithms. Our algorithm was shown to be more than 84 times faster than the previously best-known ones over the literature benchmark. Moreover, on average, it considered a number of launch and retrieval pairs that is linear on the number of customers, indicating that the algorithm's performance should be sustainable for larger instances.

Data Structures and Algorithms

What problem does this paper attempt to address?

This paper focuses on the Fixed Sequence Traveling Salesman Problem (FSTSP) with the assistance of drones. In FSTSP, a truck and a drone are used to serve a set of customers, with the drone serving one customer at a time and returning to the truck before taking off again. The objective is to minimize the time required to serve all customers and return to the warehouse.

Existing algorithms typically employ the "Sequential Primarily Remove Sequencing" approach, where a Hamiltonian cycle containing all customers is first constructed, and then some customers are removed from the cycle to be handled by the drone. The paper introduces the concept of "lazy drone properties", which suggests that for solving the FSTSP problem with a given initial cycle, it is only necessary to consider combinations of launch and recovery nodes for a subset of the drones. Based on this property, they design a new algorithm which, according to experimental results, is more than 84 times faster than previously known fastest algorithms. On average, the algorithm's consideration of launch and recovery scales logarithmically with the number of customers, indicating its sustainable performance on larger instances.

The structure of the paper includes problem definition, introduction of existing algorithms, proof of the lazy drone properties, proposal of the new "Lazy Sequencing Algorithm", as well as experimental results and comparisons. The experiments demonstrate the efficiency advantages of the new algorithm.