The Maximum Traveling Salesman Problem with Submodular Rewards

TL;DR

This paper studies the maximum reward traveling salesman problem with submodular rewards, analyzing approximation algorithms, their bounds, and practical performance, with applications in environmental monitoring using mobile sensors.

Contribution

It introduces and analyzes two approximation algorithms for the NP-hard problem with submodular rewards, providing bounds and empirical evaluation.

Findings

Second algorithm has better bounds for low curvature .

Both algorithms require (|V|^3) oracle calls.

Simulation results demonstrate practical effectiveness.

Abstract

In this paper, we look at the problem of finding the tour of maximum reward on an undirected graph where the reward is a submodular function, that has a curvature of $\kappa$, of the edges in the tour. This problem is known to be NP-hard. We analyze two simple algorithms for finding an approximate solution. Both algorithms require $O(|V|^3)$ oracle calls to the submodular function. The approximation factors are shown to be $\frac{1}{2+\kappa}$ and $\max\set{\frac{2}{3(2+\kappa)},2/3(1-\kappa)}$, respectively; so the second method has better bounds for low values of $\kappa$. We also look at how these algorithms perform for a directed graph and investigate a method to consider edge costs in addition to rewards. The problem has direct applications in monitoring an environment using autonomous mobile sensors where the sensing reward depends on the path taken. We provide simulation results to empirically evaluate the performance of the algorithms.