A man-machine approach toward solving the traveling salesman problem

The traveling salesman problem, (form a circuit through N points with no subloops in such a way as to minimize the length of the circuit), is a close kin to many board wiring problems. It has been attacked by many mathematical methods with only meager results. Only for special forms of the problem or for problems with relatively few points can it be solved exactly even using very large amounts of computer time. Heuristic procedures have been proposed and tested with only slightly better results. This paper will describe a computer-aided heuristic technique which uses only a modest amount of computer time in real time to solve large (100-200) point problems. This technique takes advantage of both the computer's and the human's problem solving abilities. The computer is not asked to solve the problem in a brute force way as is the case in many of today's heuristics but it is asked to organize the data for the human in a fashion that allows the human to solve the problem easily. The techniques employed in this paper require that the computer and the human cooperate to find the solution to the problem in reasonable amounts of both of their times. The computer initially uses a series of heuristics that produce groups of points and some partial connections of these points. The human is asked to connect the points within the groups and then connect the groups in a manner that produces a circuit and appears to the human to maximize the ratio of enclosed area to perimeter. The computer takes this solution and uses another set of heuristics to make improvements. The solution is displayed to the human and if he is satisfied the procedure stops; if not the former procedure is repeated until the human is satisfied that cost for finding a better solution exceeds his estimate of the best possible improvement that could be obtained by further work. The heuristic procedures seek to group points around information obtained from solving a series of mathematical programming problems (assignment problems) and some observed correlations between these problems and the traveling salesman problem. The results are very good. The man-machine interaction solution for all problems in the literature is within one per cent of the solution for a fraction of the computer time. The experience to date indicates that the technique can be taught to inexperienced persons and their results, after training, are similar to those of the authors'.