[Topic 3] Planning and Scheduling based on Intelligent Agents.

(1)Our research focuses on Soar architecture with a learning function. Soar is an architecture for a system to be able to work on the full range of tasks, from highly routine to extremely difficult open-ended problems; and employ the full range of problem solving methods and representations required for these tasks. A problem space is defined with symbols. Though this is possible for Soar to make a plan in the abstract level, it is difficult to adapt itself to the real world.
We propose a rule generator based on Reinforcement Learning in Soar. Reinforcement Learning is a machine learning method which addresses how an autonomous agent can learn long-term successful behavior through interaction with its environment. That differs from the more well-studied problem of supervised learning, in that the learner is not given input-output samples of the desired behavior. The learner is only supplied scalar feedback regarding the appropriateness of the actions, after they have been carried out. Our goal is to produce new production rules from a reinforcement learning function and apply them to Soar.

(2)Distributed Valued Constraint Satisfaction Algorithm and its application
As application of Distributed Valued Constraint Satisfaction Algorithm, we implement the nursing staff scheduling system by multi-agent. In this system, the event allocated on a calendar is considered as a variable, and the conditions, which should be satisfied by two or more events and in a certain specific event, are considered as constraints. And weights are given as importance of each event, then we consider the problem as Distributed Valued Constraint Satisfaction Problem(DVCSP).
As the concrete example of application of this systemCWe consider scheduling of nursing staffs in a hospital. Each event takes three values of a morning shift, an evening shift, and night shift. and event has the nurse in charge as a parameter. As constraints at this case, legal regulations, such as work of continuation, individual frame for each person such as requests of nurses, etc. can be considered. About the individual frame for each person, each user can make private constraints reflect by formulating problem as DVCSP. Of course, the information on constraints is not exhibited in this case.


Shintani lab. 2003