Releasing source locating based on Multi-Agent Reinforcement Learning with reward function designed by maximum entropy

• Published in: 2022 41st Chinese Control Conference (CCC) pp. 4688 - 4693
• Main Authors: Wang, Zhi-Pu, Zeng, Guang-Rong, Deng, Lie-Wei, Cao, Wang, Guo, Yao
• Format: Conference Proceeding
• Language: English
• Published: Technical Committee on Control Theory, Chinese Association of Automation 25.07.2022
• Subjects: Data models; Entropy; Information entropy; Maximum entropy; Multi-Agent Reinforcement Learning; Reinforcement learning; Releasing source locating; Sensors
• ISSN: 2161-2927
• DOI: 10.23919/CCC55666.2022.9902336

This paper is focused on locating the actual releasing source in the environment of multiple disturbance sources. The actual releasing source is located with multiple mobile sensors. In an attempt to avoid mobile sensors falling into the disturbance releasing source and gather at the actual releasing source quickly, an improved Multi-Agent Reinforcement Learning (MARL) with novel designed reward function is applied to guide the movement of mobile sensors. To ensure finding the actual releasing source with maximum releasing concentration, the reward function is designed based on maximum entropy (ME). Finally, MARL with reward function designed by ME and normal MARL are simulated and compared to verify the efficiency and advantage of this method.