Non-Euclidean Traveling Salesman Problem

Abstract

The traveling salesman problem (TSP) is usually studied on a Euclidean plane. When obstacles are placed on the plane, the distances are no longer Euclidean, but they still satisfy the metric axioms. Three experiments are reported in which subjects were tested on the TSP and on the shortest-path problem with obstacles. When the obstacles were simple, and they did not change the global structure of the problem, the subjects were able to produce near-optimal solutions, but the complexity of the mental mechanisms was higher than in the case of the Euclidean TSP. When obstacles were complex and changed the problem's global structure, the solutions were no longer near-optimal. Several computational models are proposed that can account for the psychophysical results.