Keywords:
Multiple Robots, Cooperation, Reinforcement Learning

Abstract:
Note: This is a practice to my PhD Proposal talk I may not be able to give much detail in the 20-min talk but I will focus mainly on the motivation and idea.

One important advantage of multirobot systems over a single robot is their performance. Multiple robots can share tasks and help each other to accomplish a mission more efficiently than a single robot. We say roughly that those robots cooperate with each other to accomplish that mission. However, "cooperation" has only been loosely defined. In some cases, multirobot performance may be worse than that of a single robot and we cannot say that those robots cooperate.

There is no theory to judge that robots cooperate or not. I will present a standard to judge this based on performance of the system. Because cooperation is preferable, my research is about finding a supportive environment for a group of robots to exhibit cooperation behavior according to my standard. I use reinforcement learning as a tool to bring about cooperation and as an indicator to the desirable enviroment for cooperation.