LEE- Kontrol ve Otomasyon Mühendisliği-Doktora
Bu koleksiyon için kalıcı URI
Gözat
Yazar "Turanlı, Mert" ile LEE- Kontrol ve Otomasyon Mühendisliği-Doktora'a göz atma
Sayfa başına sonuç
Sıralama Seçenekleri
-
ÖgeMulti-agent coverage control with adaptation to performance variations and imprecise localization( 2020) Turanlı, Mert ; Temeltaş, Hakan ; 637482 ; Kontrol ve Otomasyon Mühendisliği Ana Bilim DalıIn this thesis, an adaptive collaboration approach for a multi-agent system consisting of nonholonomic wheeled mobile robots is proposed. The positions of the agents are not known precisely but their locations are known to be within uncertainty circles. For the collaboration among the robots, the workspace partitioning algorithm is chosen as Guaranteed Power Voronoi Diagram (GPVD or GPD) which not only takes the localization uncertainty into account but also is capable of changing the regions of the generator points with respect to corresponding weight parameters. Also, the assumption is that the actuation capabilities of the robots are different from each other. The agents do not know those parameters related to their actuation performances beforehand. The contribution of the thesis is that the performance parameters of the agents are learned online by the proposed adaptive estimator algorithm and Hopfield Neural Network (HNN) estimator under localization uncertainty. The proposed algorithm is based on the coverage control which performs collaboration among the robots by assigning the regions from the workspace according to their actuation performances automatically. The definition of the actuation performances is different capabilities of the agents. The examples of strong actuation performances may include powerful motors and favorable terrain while wheel slip and weak motors can be counted as examples for the weak actuator performances. The proposed multi-agent collaborative coverage algorithm learns the performance parameters of the robots by using two approaches proposed in the thesis. The first approach is based on an adaptive estimator with a non-holonomic estimation model. The second method uses an HNN estimator. The theoretical proof, analysis and verification of the aforementioned methods are given in the related sections. After estimating the performance parameters, the weights are calculated using a neighbor based weight estimation algorithm. The weight variables are utilized in the GPD algorithm so that the workspace is partitioned according to the performance parameters of the agents in a guaranteed sense. At the end, the agents take regions from the workspace according to their actuation performances and achieve the optimal collaborative coverage so that the agents with strong actuators take larger regions from the environment than the agents with poor actuators. Thus, the collaborative coverage algorithm enables the robots to deploy themselves to an optimal configuration which minimizes the total coverage cost by taking imprecise localization into account. Moreover, a multi-agent coverage collaboration method with an energy-efficient optimal coverage control law and Hopfield networks is proposed in the related section. By using the algorithm a trade-off between coverage time and energy consumption among agents can be done. Meanwhile, the collaboration is achieved according to the actuation performances of the agents. The theoretical results are verified with MATLAB and ROS/Gazebo simulations and experiments that show the efficiency of the algorithm. The ROS implementation of the algorithm is explained. The experimental results are given in the related section.