多目標進化式演算法於四足仿生機器人故障時之自動調適行為

Automatic Adaptive Locomotion on Broken-down Quadrupedal Biomorphic Robot by Multi-Objective Evolutionary Algorithms

陳厚諭(Hou-Yu Chen)

指導教授 ： 吳世弘

朝陽科技大學/資訊學院/資訊工程系/碩士(2014年)

摘要

近年來機器人漸漸融入於我們的生活中，傳統機器人的行為是依賴一套動力學指令來控制它，當周遭環境改變時，需要編制一套新的行為來應付變動中的環境。進化式機器人可以彌補傳統機器人需要人重新編制的缺點，透過進化式演算法立即編制一套適合目前環境的行為，這就是進化式機器人的優點。
在機器人學中進化式機器人一直是個重要的研究發展議題，主要是由機器人和機器人所在的環境兩大因素所構成，而研究的目標，是要讓進化式機器人能夠自主開發一套行為來達成指派的任務，本篇碩論比較多種進化式演算法(基因演算法、多目標隔代交替基因演算法、仿生機器人瀰集演算法)，運作在四足、四足多關節以及部分故障的仿生機器人上。當指派任務給機器人，機器人如果只針對單一指標演化，是不夠全面的，所以我們嘗試採用多目標演化。在我們實驗中，會在終點處設置一個類似燈塔的光源當作目標，使機器人能夠快速且平穩朝向目標移動，作為我們實驗的主要目的。從我們實驗結果可以證明，不論是四足或四足多關節甚至四足機器人面臨到馬達故障情況，可以經由進化式演算法立即調適進而改善機器人順利達成指定任務。

關鍵字

進化式機器人 ； 仿生機器人 ； 進化式演算法 ； 多目標演化

並列摘要

In recent years, the robots gradually integrated into our lives. Traditional robots locomotion relied on the kinetic analysis to design a set of instructions to control the robot. When the surrounding environment changed, human had to develop a code of conduct to deal changes in the environment. Evolutionary robot using evolutionary algorithm to solve this problem, evolutionary robot can adapt its behavior to fit the current environment immediately, this is the advantage of evolutionary robots.
Evolutionary robot research has become an interesting topic recently. Specifically, this research focuses on evolution and learning, evolution is adaptation of robots to the environment. Learning is a task-oriented process whereby the robot gains the ability to achieve a given goal in the environment. We use evolutionary algorithms like GA algorithm, MOIEGA (Multi-objective intergeneration exchange genetic algorithms), BRMA(Biomorphic Robot Memetic algorithm) to control the robot. Our biomorphic robots have four legs and each leg has several joints. We also test on a partially break down robot. When assigning task to the robot, the robot has to evolve to fit several indexes, so our study is a multi-objective evolutionary robotics.
In our experiments, we set up a beacon light as a goal and the robot evolves to move quickly and smoothly toward the goal. We adopt online evolutionary algorithms and test them on the quadrupedal robot. The experimental results show that the robot can adjust its actions from totally random behaviors to move toward the goal quickly and smoothly.

並列關鍵字

Evolution robotics ； Biomorphic robot ； Evolutionary algorithm ； Multi-objective evolution

參考文獻

[1]Dilip Kumar Pratihar ,“Evolutionary robotics–A review, ”SADHANA Volume 28, Part 6, pp.999-1009, (2003).

[4]Andrew L.Nelson, Gregory J.Barlow, Lefter Doitsidis,“Fitness functions in evolutionary robotics: A survey andanalys is”, Robotics and Autonomous System 57, pp. 345-370,(2009).

[7]Shi, Y. and R.C. Eberhart "Empirical study of particle swarm optimization. "Proceedings of the IEEE Congress onEvolutionary Computation(CEC), Washington, DC, (1999).

[12]S. Nolfi and D. Floreano, "Co-evolving predator and prey robots: Do ‘arms races ’arise in artificial evolution? "Artificial Life, Vol. 4, pp.311-335, (1998).

[14]N. J. Rodriguez, J. H. Garzon, and J. J. Soriano, “Morphology, kinematics and stability of a Scarabaeus as bases for design of robots,” in Proceedings of the Electronics, Robotics and Automotive Mechanics Conference, Vol. 2, pp.151-156, (2006).