Abstract
Earlier publications about learning control systems used adaptive critic schemes. We argue that the results published about adaptive critic schemes in control problems appeared positive due to the choice of the inverted pendulum or cart-pole system as a test case. This conclusion was drawn after e xamination of a number of publications and after our experiments with the original adaptive critic scheme. In this paper we describe an adaptive critic scheme derived from a control point of view. This scheme allows us to indicate more clearly the limitations of the scheme in a control context. We show how an adaptive critic scheme fails to learn a set-point control for a simple second order non-linear process in a simulation.
Original language | English |
---|---|
Title of host publication | ICANN ’93 |
Subtitle of host publication | Proceedings of the International Conference on Artificial Neural Networks Amsterdam, The Netherlands 13–16 September 1993 |
Editors | Stan Gielen, Bert Kappen |
Place of Publication | London |
Publisher | Springer |
Pages | 810-813 |
Number of pages | 4 |
ISBN (Electronic) | 978-1-4471-2063-6 |
ISBN (Print) | 978-3-540-19839-0 |
DOIs | |
Publication status | Published - 13 Sept 1993 |
Event | International Conference on Artificial Neural Networks, ICANN 1993 - Amsterdam, Netherlands Duration: 13 Sept 1993 → 16 Sept 1993 |
Conference
Conference | International Conference on Artificial Neural Networks, ICANN 1993 |
---|---|
Abbreviated title | ICANN |
Country/Territory | Netherlands |
City | Amsterdam |
Period | 13/09/93 → 16/09/93 |
Keywords
- Inverted pendulum
- Criterion surface
- Flat region
- State feedback controller
- Criterion predictor