Abstract
A computing architecture for robot adaptive control called a neuromorphic architecture is described. It is based on recently computational features of nonlinear neural networks. These features are the associative storage and retrieval of knowledge and the uniform (i. e. independent of the network's dimension) rate of convergence of the network's dynamics towards steady states. The major benefits expected from this architecture are: (1) adaptation rate for finding optimal parameters that is faster than model reference adaptive control or self-tuning regulator methods; (2) simpler controller structure for multiparameter adaptive control (i. e. , the complexity of controller does not grow exponentially with the number of unknown parameters); and (3) adaptation over both discrete- and continuous-parameter domains.
Original language | English (US) |
---|---|
Pages | iv/567-572 |
State | Published - 1987 |
Externally published | Yes |
All Science Journal Classification (ASJC) codes
- General Engineering