| Project/Area Number |
08650497
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
計測・制御工学
|
|---|
| Research Institution | Kyushu University |
|---|
Principal Investigator |
ZHA Hongbin Associate Professor, Department of Intelligent Systems, Graduate School of Information Science & Electrical Eng., Kyushu University, 大学院・システム情報科学研究科, 助教授 (80225680)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
OKADA Nobuhiro Assistant Professor, University Computation Center, Kyushu University, 工学部, 講師 (80224020)
HASEGAWA Tsutomu Professor, Department of Intelligent Systems, Graduate School of Information Sci, 大学院・システム情報科学研究科, 教授 (00243890)
|
|---|
| Project Period (FY) |
1996 – 1998
|
| Project Status |
Completed (Fiscal Year 1998)
|
| Budget Amount *help |
¥2,400,000 (Direct Cost: ¥2,400,000)
Fiscal Year 1998: ¥200,000 (Direct Cost: ¥200,000)
Fiscal Year 1997: ¥900,000 (Direct Cost: ¥900,000)
Fiscal Year 1996: ¥1,300,000 (Direct Cost: ¥1,300,000)
|
|---|
| Keywords | visual-motor coordination / machine learning / self-organization / multi-robot systems / neural networks / ニューラルネットワーク |
|---|
| Research Abstract |
The research aims to develop new methods of coordinating different functions in a robot system to accomplish complex tasks. Two kinds of functions are taken into consideration here : 1) visual sensing ; 2) motion control. They are coordinated mainly by making the robot automatically learn the motion trajectories when the target is recognized by the visual sensor. Basic idea hehind the learning process is to use self-organization principles in neural computing to simplify the originally complicated nonlinear learning problem into locally linear subproblems. The research was carried out in three stages, and main results are given in the following.
1) Learning algorithms for off-line simulators : Such an algorithm for learning target reaching by a six DOF manipulator is developed. It is implemented in a simple simulator that is able to simulate 3-D real environments with obstacles. New ideas such as potential-based optimization are utilized to deal with the complexity of environments.
2) On-line coordination using real manipulators : We built a real system composed of a high-quality workstation, a stereo-camera sensor and a six DOF manipulator. The learning algorithm developed in the first stage was realized in the system with special efforts for implementation efficiency. At the same time, we also solved the problem of obstacle avoidance when there is an object between the robot and the target.
3) Learning in tasks using multiple robots : The above algorithms were extended and applied to multi-robot systems. Two kinds of tasks based on the systems are investigated : 1) pushing an object uoward on a slanted plane ; 2) object rotation by a multi-finger hand, In these tasks, visual sensor are used to provide information on the system configuration as well as the states of objects under manipulation. It was shown that the visuo-motor coordination highly benefited such tasks of greater complexity and more uncertainties.
|
