2006 Fiscal Year Final Research Report Summary
A study on topology and weight evolving artificial neural networks for co-evolutionary multi-robot systems
| Project/Area Number |
17500144
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Sensitivity informatics/Soft computing
|
|---|
| Research Institution | HIROSHIMA UNIVERSITY (2006)
Kobe University (2005) |
|---|
Principal Investigator |
OHKURA Kazuhiro Hiroshima University, Graduate School of Engineering, Professor, 大学院工学研究科, 教授 (40252788)
|
|---|
| Project Period (FY) |
2005 – 2006
|
| Keywords | soft computing / intelligent robotics / agent / artificial intelligence / intelligent machines / artificial neural networks / artificial evolution / computational grid |
|---|
| Research Abstract |
There have been various methods proposed for realizing intelligence on artifacts. Among them, the approach called artificial life (a.k.a. A-Life) proposed by C. Langton in 1987 attracted a lot of attention in 1990's. After the enthusiastic movement toward A-Life, the research field called evolutionary robotics has been emerging as an approach to building the artificial brain for autonomous robots. So far, although evolutionary robotics has got primary successful results, the principal researcher considers that there are at least three major problems to break through the current status to the next stage in this field. To each of the three problems, the principal researcher has shown the following results.
I. The traditional theoretical framework of evolutionary computation cannot be applied to evolutionary robotics.
To this problem, the principal researcher asserts that ER needs the theory of open-ended artificial evolution that is fundamentally different from the traditional artificial e
… More
volution in the sense that the convergence in genotype implies the end of genetic search. The principal researcher contributed to building the open-ended evolutionary algorithms that have the self-adaptive control mechanisms of the practical genetic mutation rates and the natural selection pressure.
II. It is practically impossible to remove the redundancy from genotype that represents topology and weight evolving artificial neural networks to expect the effective genetic search.
To this problem, the principal researcher proposed a novel method called Mutation-Based Evolving Artificial Neural Networks (MBEANN). The robust performance was proved by the standard benchmark problem of double-pole balancing.
III. No framework was proposed for providing very large computational resources for evolutionary robotics.
To this problem, the principal researcher proposed the computational grid specialized for evolutionary robotics. The grid has a novel task-scheduling algorithm called R3Q. The effectiveness was examined by implementing R3Q into the WAN grid environment to conduct a series of ER experiments. Less
|
Research Products
(26 results)
