Hideaki Takeda's Publication
- A. Ueno, H. Takeda and T. Nishida: Cooperation of
Cognitive Learning and Behavior Learning, in Proceedings of the 1999
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS
99), Vol. 1, pp. 387–392 (1999).
Reinforcement learning is very useful for robots with little a
priori knowledge in acquiring appropriate behavior. Autonomous segmentation
of the continuous state space is a promising method for reinforcement
learning in real robots. This is a kind of cognitive learning. We think
cognitive learning should be based on similarity of rewards since the task of
the robot is expressed in the rewards in the general reinforcement learning
problem. Furthermore, cognitive learning should be performed in an on-line
way for flexibility to changes of the environment and immediate convergence
of leaning. This paper describes a learning system which can learn a state
representation and a behavior policy simultaneously while executing the task.
We call the system Situation Transition Network System (STNS). As cognitive
learning, it extracts situations and maintains them dynamically in the
continuous state space on the basis of rewards from the environment. As
behavior learning, it makes an MDP model of environment and performs partial
planning on the model. This is a kind of reinforcement learning. The results
of computer simulations are given.
Hideaki Takeda (National Institute of Informatics)