20.500.12556/RUL-121804 Dodeljevanje transportnih naročil na osnovi vzpodbujevalnega učenja v sistemu mobilnih robotov Reinforcement learning-based task assignment in a mobile robot system Moderne rešitve na področju logistike obsegajo uporabo sistemov z več mobilnimi roboti. Pogoj za uspešno implementacijo takšnega sistema je učinkovit sistem dodeljevanja transportnih naročil. Ta mora skupini robotov transportna naročila dodeljevati tako, da jih ta opravijo čim več. Problem smo prepoznali kot problem optimizacije več kriterijev. Cilj naloge je razvoj algoritma, ki problem dodeljevanja rešuje na osnovi vzpodbujevalnega učenja. Algoritem smo razvijali v programskem okolju ROS. Razvili smo dodeljevalni algoritem s poudarkom na hitremu učenju. Razviti algoritem smo preizkusili v simuliranemu preizkusu. Njegovo delovanje smo primerjali z algoritmi, ki naročila dodeljujejo na osnovi preprostih pravil in izpolnjujejo le en kriterij problema. Vsak način dodeljevanja smo preizkušali v preizkusu, ki je trajal eno uro. Pri dodeljevanju z razvitim algoritmom so roboti zaključili največje število transportnih naročil. Meritve prepotovanih razdalj in časov opravljanja nalog so potrdile enostranskost preprostih pravil ter večkriterijski proces odločanja razvitega algoritma. Modern logistic solutions encompass the use of mobile robot systems. To achieve a successful implementation of such a system, one must consider an efficient design of transportation task assignment system. Main responsibility of a task assignment system is to allocate tasks in such manner that as many tasks get completed in a given time frame. This problem is recognized as a multicriteria optimization problem. The purpose of this thesis is to develop a task assignment algorithm that is based on reinforcement learning. The proposed algorithm was developed using ROS platform. We developed an algorithm with an emphasis on fast learning. The proposed algorithm was tested in a simulated environment. It was tested alongside simple task assignment rules that meet only single criterion of an assignment problem. Every task assignment algorithm was tested in an hour long experiment. Robots managed to complete the highest number of tasks in the case of the developed solution. Measurements of traveled distances and task completion times confirmed the one-sided decisions of simple rules, and the multicriteria decision-making process of the developed algorithm. robotika optimizacija vzpodbujevalno učenje dodeljevanje naročil sistem mobilnih robotov ROS platforma robotics optimization reinforcement learning task assignment mobile robot system ROS platform true false false [M. Knap] Slovenski jezik Angleški jezik Magistrsko delo/naloga 2020-10-30 08:30:45 2020-10-30 08:30:50 2022-09-02 04:01:06 0000-00-00 00:00:00 2020 0 Ljubljana 0 XXIV, 75 str. 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 007.52:519.8:004.8(043.2) 225977 38621699 6430.pdf 6430.pdf 1 DCDE572D5307DD56224890BAA1621232 083d077098aff4361aeffd1229278b0d37595ed1c170e612dc9e7f781864b2e1 fb34c318-a1ba-11eb-a523-00155dcfd717 https://repozitorij.uni-lj.si/Dokument.php?lang=slv&id=136816 Fakulteta za strojništvo 0 0 0