| 103D5 |
|---|
| Multi-agent reinforcement learning with marking to adapt to dynamical environment and improve efficiency |
| AUTHORS |
| Tomohiro Fujiki, Junichi Murata (Kyushu University) Kotaro Hirasawa (Waseda University) |
| ABSTRACT |
| When there are many agents in an environment, the environment viewed from an agent changes dynamically by the actions of other agent. If the agents cannot perceive the changes of the environment, the agents with reinforcement learning cannot get good action-rules or enough efficiency of the system. In this paper, it is proposed that the agent "marks" the parts changed in the environment, so that the agent can remember the change of the environment. Also other agents perceive the mark that the agent has left. In this way, the agents have an easy communication media. The multi-agent systems gets better efficiency by the mark. In addition, the agents learn the effect of the mark and decide whether or not they use the mark depending on the situation. Also the mechanism of the mark is studied in a complex environment. |
Back to Program
Time Table
Top of This Site
Annual Conference 2002