一般注記Due to the exibility of multi-legged robots, they can be applied in several applications in which regular wheel-based robots are not suitable to perform. However, the multi-legged robots require a number of sensors and actuators. In a normal case, legged robots can perform properly with a controller programmed by the user, but there are some failures occurring when some sensors and actuators are no longer available. The legged robots are nonfunctional after getting damaged since the prior control strategies cannot be employed to operate e ciently with transferred models. A self-recovery method can overcome this problem by nding alternative behavior of the robot. Although these techniques make robots possible to walk after damaging, the existing methods require a lot of time to operate. This research mainly focuses on developing the algorithm for self-recovery, concerning about the learning time and complexities. Not only the self-recovery method is studied in this research, but also other factors that can help the robot to be able to move again. The novel structure model of the quadruped robot has been proposed in this study. The caterpillar-inspired quadruped robot (CIQR) is developed to imitate the caterpillar crawling locomotion when the robot has a small number of active legs. The caterpillars' proleg is added on the robot limb to improve the ability to move after some parts of the robot got damaged. This lets the legged robot become movable even if it has only one leg. However, the structure of legs has to be designed circumspectly due to the face that the proleg can limit the reachable space of robots leg while operating with normal quadruped gaits. In this paper, the new shape of the robotic leg is designed with the inspiration of caterpillar and optimized using PSO algorithm. The tness function of PSO is set as the distance that the robot can travel in both crawling and trotting gait. To discover the new model, the PSO-based Leg-loss Identi cation method (PLI) is proposed. The PLI method uses only onboard sensors that let the robot become more versatile. Particle swarm optimization is utilized to optimize the tness function that is set as the resemblance of candidate models and actually damaged the robot. The acoustic-based fault diagnosis for legged robots (AFL) is developed to detect the abnormalities of joints. The sound of servo motors is recorded simultaneously while a multi-legged robot is executed to perform speci c actions. The results show that both proposed methods can detect the fault parts properly with the broken robot in the experiments. Moreover, the development of new bio-inspired locomotion method is conducted to help the legged robot that has a small number of legs to be able to move again after getting damage. The concept is based on the movement of mudskipper in nature. Self-learning mudskipper-inspired crawling method (SLMIC) is proposed in this study. The reinforcement learning method, Q-learning, is integrated to improve the adaptability of locomotion. The results show that the proposed method is feasible to employ a damaged robot that has two legs. According to the results of proposed works, the upshot of this study is the possibility that the methodology used in this study is feasible to implement with the damaged legged robot in a practical situation.
(主査) 教授 小林 幸徳, 教授 梶原 逸朗, 教授 近野 敦(情報科学研究院)
工学院(人間機械システムデザイン専攻)
コレクション(個別)国立国会図書館デジタルコレクション > デジタル化資料 > 博士論文
受理日(W3CDTF)2019-08-03T12:23:27+09:00
連携機関・データベース国立国会図書館 : 国立国会図書館デジタルコレクション