| Project/Area Number |
10680408
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
情報システム学(含情報図書館学)
|
|---|
| Research Institution | Kyushu Institute of Technology |
|---|
Principal Investigator |
EJIMA Toshiaki Department of Computer Science, Kyushu Institute of Technology, Professor, 情報工学部, 教授 (00124553)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
OHASHI Takeshi Department of Computer Science, Kyushu Institute of Technology, Ass. Professor, 情報工学部, 助手 (00233239)
YOSHIDA Takaichi Department of Computer Science, Kyushu Institute of Technology, Ass. Professor, 情報工学部, 助教授 (70200996)
|
|---|
| Project Period (FY) |
1998 – 2000
|
| Project Status |
Completed (Fiscal Year 1999)
|
| Budget Amount *help |
¥3,300,000 (Direct Cost: ¥3,300,000)
Fiscal Year 1999: ¥700,000 (Direct Cost: ¥700,000)
Fiscal Year 1998: ¥2,600,000 (Direct Cost: ¥2,600,000)
|
|---|
| Keywords | reinforcement / Learning / autonomous robot / behavior Acquition / Synceses / Q-Learning / Q-学習 / 自律型ロボット / 自己組織化 |
|---|
| Research Abstract |
Reinforcement learning has been used as a method that is useful for an autonomous robot to select an appropriate action in each state with little or no premise knowledge. Typically, even if an autonomous robot has continuous sensor values, sensor space is quantized to reduce learning time. However, the reinforcement learning algorithms including Q-learning suffer from errors due to state space sampling.
To overcome the above, we propose Extended Q-learning (EQ-learning) based on Q-learning creates mapping that maps a continuous sensor space to a descrete action space. Through EQ-learning, action-value function approximation is represented by a summation of weighted base functions, and autonomous robot adjusts only weight of base function by robot learning. In order to obtain a simpler learning model, other parameters are calculated automatically by unification of two similar base functions.
|
