http://www.cnr.it/ontology/cnr/individuo/prodotto/ID95124
Adaptive Q-learning algorithm with constant learning rate parameters (Contributo in atti di convegno)
- Type
- Label
- Adaptive Q-learning algorithm with constant learning rate parameters (Contributo in atti di convegno) (literal)
- Anno
- 2001-01-01T00:00:00+01:00 (literal)
- Alternative label
D'Orazio T. , Cicirelli G. , Distante A. (2001)
Adaptive Q-learning algorithm with constant learning rate parameters
in Intelligent Systems and Control, Tampa, Florida, 19-22 Novembre 2001
(literal)
- Http://www.cnr.it/ontology/cnr/pubblicazioni.owl#autori
- D'Orazio T. , Cicirelli G. , Distante A. (literal)
- Http://www.cnr.it/ontology/cnr/pubblicazioni.owl#altreInformazioni
- The product is publisehd in the Proc. of ISC 2001 which is organized by the
IASTED society. (literal)
- Http://www.cnr.it/ontology/cnr/pubblicazioni.owl#descrizioneSinteticaDelProdotto
- In robotic tasks adaptive learning systems are required as soon as a
change of behavior is necessary to manage new situations.
In this paper we compare the convergence rate of Q-values to the optimal
ones considering different learning-rate parameters (constant and
decreasing ones). In particular we suggest a weighted updating of the
Q-values that allows both the adaptivity to new situation and the smoothing
of noise. The used experimental test-bed is a grid-world domain where a
simulated agent learns to reach a particular goal state. After some
iterations the goal position in the environment is moved and the algorithm
is able to learn a new policy to accomplish the task.
(literal)
- Http://www.cnr.it/ontology/cnr/pubblicazioni.owl#affiliazioni
- Titolo
- Adaptive Q-learning algorithm with constant learning rate parameters (literal)
- Prodotto di
- Autore CNR
- Insieme di parole chiave
Incoming links:
- Prodotto
- Autore CNR di
- Insieme di parole chiave di