Abstract | ||
---|---|---|
The present paper discusses an optimal control method of biological robot arm which has re- dundancy of the mapping from the control input to the task goal. The control input space is divided into a cou- ple of subspaces according to a priority order depending on the progress and stability of learning. In the pro- posed method, the search noise which is required for reinforcement learning is restricted within the rst pri- ority subspace. Then the constraint is relaxed with the progress of learning, and the search space extends to the second priority subspace in accordance with the his- tory of learning. The method was applied to the mus- culoskeletal system as an example of biological control systems. Dynamic manipulation is obtained through reinforcement learning with no previous knowledge of the arm's dynamics. The eectiv eness of the proposed method is shown by computational simulation. |
Year | DOI | Venue |
---|---|---|
2002 | 10.1109/SICE.2002.1195433 | SICE 2002. Proceedings of the 41st SICE Annual Conference |
Keywords | DocType | Volume |
neural network,manipulator dynamics,optimal control,motion control,reinforcement learning,learning (artificial intelligence),musculoskeletal model,dynamics,dynamic manipulation,musculoskeletal system,biomimetic robot,over-actuated system,search noise,physiological models,redundant manipulators,biological robot arm,impedance adjustment,search space,stability,biomimetic learning control system,neural nets,learning artificial intelligence | Conference | 1 |
ISBN | Citations | PageRank |
0-7803-7631-5 | 6 | 0.55 |
References | Authors | |
3 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Jun Izawa | 1 | 57 | 5.47 |
Toshiyuki Kondo | 2 | 131 | 28.57 |
Koji Ito | 3 | 24 | 7.23 |