In this paper, we examine the effectiveness of multimodal genetic programming (MMGP) on the wall-following problem, which is a well-known benchmark problem of genetic programming (GP). MMGP aims to obtain multiple local optimal programs, including global optimal programs, that is, programs that achieve the same goal with different program structures. In this paper, we apply MMGP to the wall-following problem. The purpose of the wall-following problem is to find a program to control a robot having twelve distance sensors and four movements to follow irregular walls. We expect that there are several local optimal programs in the wall-following problem, which use different combinations of sensors. An experiment is conducted to investigate whether MMGP can get local optimal programs simultaneously for the wall-following problem. This experiment compares MMGP with a simple GP. The experimental results reveal that MMGP can achieve higher acquisition ratio of local optimal programs than the simple GP.
[1]
Anthony K. H. Tung,et al.
Similarity evaluation on tree-structured data
,
2005,
SIGMOD '05.
[2]
John R. Koza,et al.
Genetic programming as a means for programming computers by natural selection
,
1994
.
[3]
Ruck Thawonmas,et al.
A Study on Multimodal Genetic Programming Introducing Program Simplification
,
2018,
2018 Joint 10th International Conference on Soft Computing and Intelligent Systems (SCIS) and 19th International Symposium on Advanced Intelligent Systems (ISIS).
[4]
Ruck Thawonmas,et al.
Multimodal genetic programming by using tree structure similarity clustering
,
2017,
2017 IEEE 10th International Workshop on Computational Intelligence and Applications (IWCIA).