Proceedings of the seventh international conference (1990) on Machine learning
Technical Note: \cal Q-Learning
Machine Learning
Introduction to Reinforcement Learning
Introduction to Reinforcement Learning
| Hi-index | 0.00 |
Previously, we proposed new Q-learning based on CMAC[6]. The target filed is made from these objects: each agent, exit points and walls. Agents select random action about first object, which come up a view area of agent. But this system is using CMAC system and added new rules will keep work with the behavior on similar view. The proposed learning system has a multi-layer style rule base. Those several layers have different members of the object. If one rule base has no valid rule, other rule is able to output the incomplete actions by slightly valid objects of the rule base. However those actions are not random walking. In any cases, agents should take a behavior by the rule system. So if the agent receives penalty about a wrong action, the Q-learning system makes change the selection weights in the current layer. The proposed Q-learning system continues learning and adds the new object and changes the next layer when the agents has observed new objects. In dynamic cases, agents should work any cases, but old rules as often as dose not work a current state, because new objects are changed any situations for whole agents, dynamically. We should research the relationships between current rules and old rules and discuss the reasonable interrelation level about those rules. In this paper, we will discuss our proposed idea and show results on the fire panic problem.