多機器人合作架構及其在機器人足球比賽之應用

Abstract

雖然多機器人合作的問題在過去十幾年間已被廣泛的研究過，但針對機器人團隊競爭這個題目所做的討論卻不多見。在這篇論文裡，我們提出一套稱為Emerge的分散式軟體架構。這個軟體架構可以讓一個團隊內的機器人有效地相互合作，以與另外一隻機器人隊伍互相競爭。藉由Emerge這個架構，團隊內的每一隻機器人皆可針對任務的需求、環境的變化、敵我間的互動、幾器人內部的狀態與雙方相對的表現做有效且可靠的行為選擇。為了達成這樣的行為選擇機制，我們在Emerge中提出了四種不同的內在動機，分別為practicability、feasibility、tendency與execution priority。藉由這四種以數字表示的內在動機，我們可以很方便地將人類關於某種團隊競爭的知識融入Emerge中，即使這些知識裡面含有一些與電腦的精確不太相稱的模糊概念。此外，Emerge本身還具有學習的能力，可以使這隻機器人團隊由過去的任務中吸取經驗，改善團隊的整體表現。這樣的學習動作主要是藉由基因演算法動態調整Emerge中的內在動機，改變每一隻機器人的行為選擇機制，以期達成更好的團隊表現。最後，我們藉由模擬與實際的三對三機器人足球賽驗證了這個軟體架構可行性。

The research of multiagent systems has been widely studied for many years, however, the topic of multirobot team competitions has not been extensively investigated. In this thesis, we propose a fully distributed, behavior-based software architecture termed Emerge, which facilitates flexible and reliable coordination on a team of robots performing tasks that may be subverted by another team of robots. Emerge enables the robots to achieve flexible action selections in response to the requirements of the task, the current environmental situation, the activities of the teammates and opponents, the internal states of the robot, and the performance measures inhered in the competitions. This reliable action selection process is achieved through the use of four mathematically-modeled motivations: practicability, feasibility, tendency and execution priority. With the aid of these motivations, the human knowledge, which usually involves a number of fuzzy concepts, may be conveniently merged into the robot team by following a clear and logical design procedure. Moreover, Emerge also incorporates the capability of learning. Through the use of genetic algorithm, the robot team may learn from the past task execution experiences and improve its performance. The feasibility of this architecture has been demonstrated through simulations and experiments on a team of mobile robots performing 3-on-3 robot soccer games.