Application of reinforcement learning to balancing of Acrobot
Authors:
- J. YoshimotoJYJ. Yoshimoto
- S. IshiiSIS. Ishii
- M. SatoMSM. Sato
- 20 January 2003
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 5 (1062922X) , 516-521
Abstract
The Acrobot is a two-link robot, actuated only at the joint between the two links. It is one of difficult tasks in reinforcement learning (RL) to control the Acrobot because it has nonlinear dynamics and continuous state and action spaces. In this article, we discuss applying the RL to the task of balancing control of the Acrobot. Our RL method has an architecture similar to the actor-critic. The actor and the critic are approximated by normalized Gaussian networks, which are trained by an online EM algorithm. We also introduce eligibility traces for our actor-critic architecture. Our computer simulation shows that our method is able to achieve fairly good control with a small number of trials.Keywords
This publication has 5 references indexed in Scilit:
- On-line EM Algorithm for the Normalized Gaussian NetworkNeural Computation, 2000
- Practical issues in temporal difference learningMachine Learning, 1992
- Nonlinear controllers for non-integrable systems: the Acrobot example1990
- Fast Learning in Networks of Locally-Tuned Processing UnitsNeural Computation, 1989
- Neuronlike adaptive elements that can solve difficult learning control problemsIEEE Transactions on Systems, Man, and Cybernetics, 1983