Improvement of Learning Surface by Carefully-designed Interaction between Control and Mechanical Systems
| Project/Area Number |
16560382
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Control engineering
|
|---|
| Research Institution | Nagoya University |
|---|
Principal Investigator |
ISHIGURO Akio Nagoya University, Dept.of Computational Science and Eng., Associate Professor, 大学院・工学研究科, 助教授 (90232280)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
KAWAKATSU Toshihiro Tohoku University, Dept.of Physics, Professor, 大学院・理学研究科, 教授 (20214596)
|
|---|
| Project Period (FY) |
2004 – 2005
|
| Project Status |
Completed (Fiscal Year 2005)
|
| Budget Amount *help |
¥3,600,000 (Direct Cost: ¥3,600,000)
Fiscal Year 2005: ¥1,500,000 (Direct Cost: ¥1,500,000)
Fiscal Year 2004: ¥2,100,000 (Direct Cost: ¥2,100,000)
|
|---|
| Keywords | Emergence / Real-time learning / Autonomous decentralized Control / Protein folding / Improvement of learning surface / Well-balanced coupling |
|---|
| Research Abstract |
In the field of robotics, traditionally, a so-called "hardware first, software last" based design approach has been employed and thus the interplay between control and mechanical systems has not been investigated in depth to date. Recently, however, it has been widely accepted that the emergence of intelligence is strongly influenced by not only control systems but also embodiments, which is the physical properties of a robots' body. In light of these facts, this study intends to deal with the interdependency between control and mechanical systems, particularly from the viewpoint of learning. More specifically, the aim of this study is to clearly answer the following questions : (1) how should control and mechanical systems be coupled? ; and (2) what sort of interesting phenomena will emerge under such a well-balanced coupling. To this end, a decentralized control of a snake-like robot consisting of multiple body segments was mainly employed as a practical example. Despite its simplicity, the obtained results strongly suggest that there exist a best coupling between the control and body dynamics, which has never been clearly mentioned so far.
This approach is totally different from the concept of the recent "learning theories" where people accept singularities in the learning surface and search for a high-functional and sophisticated algorithm. If we take advantage of such degrees of freedom in the body dynamics, we can modify the landscape of the learning surface so that it does not have a singularity. This allows us to expect that even a "cheap" algorithm can satisfy the requirements for the finite-time convergence and the robustness.
|
Report
(3 results)
Research Products
(21 results)
