| 講演抄録/キーワード |
| 講演名 |
2022-12-21 16:30
Agent based Modeling and Reinforcement Learning for optimal allocation of resources ○Rashmi Tilak・Toshiharu Sugawara(Waseda Univ.) AI2022-45 |
| 抄録 |
(和) |
We propose a model and notation for business process for delivery of parcels using drones and attempt to improve the total efficiency of the process using reinforcement learning. Although modeling the business processing is one of important applications of multi-agent systems, it is a challenge to design and control the processing efficiently. For this purpose, we train several drones in a way that helps them determine the right number of resources keeping in view the main factors using reinforcement learning. We also examine the use of two types of Q learning algorithms --- temporal difference (TD) and SARSA --- and investigate the difference between the learned behaviors using them, such as utilization percentage of drones, queue size of packages, idle drones. We show that the trained model outperforms the model with no learning involved and SARSA results in the better performance due to their safer learning. |
| (英) |
We propose a model and notation for business process for delivery of parcels using drones and attempt to improve the total efficiency of the process using reinforcement learning. Although modeling the business processing is one of important applications of multi-agent systems, it is a challenge to design and control the processing efficiently. For this purpose, we train several drones in a way that helps them determine the right number of resources keeping in view the main factors using reinforcement learning. We also examine the use of two types of Q learning algorithms --- temporal difference (TD) and SARSA --- and investigate the difference between the learned behaviors using them, such as utilization percentage of drones, queue size of packages, idle drones. We show that the trained model outperforms the model with no learning involved and SARSA results in the better performance due to their safer learning. |
| キーワード |
(和) |
drones / SARSA / Reinforcement learning / Warehouse delivery / / / / |
| (英) |
drones / SARSA / Reinforcement learning / Warehouse delivery / / / / |
| 文献情報 |
信学技報, vol. 122, no. 322, AI2022-45, pp. 68-73, 2022年12月. |
| 資料番号 |
AI2022-45 |
| 発行日 |
2022-12-14 (AI) |
| ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
| PDFダウンロード |
AI2022-45 |