Multi Agent Deep Reinforcement Learning
Multi-agent deep reinforcement learning (MADRL) focuses on training multiple AI agents to collaborate or compete within a shared environment, aiming to optimize collective performance through learning. Current research emphasizes developing efficient algorithms like MADDPG and variations of Q-learning, often incorporating transformer networks for improved feature representation and handling complex interactions, and exploring different training paradigms such as centralized training with decentralized execution. This field is significant for its potential to solve complex real-world problems across diverse domains, including autonomous driving, robotics, network optimization, and resource management, by enabling more robust and adaptable systems.
Papers
Learning to Navigate in Turbulent Flows with Aerial Robot Swarms: A Cooperative Deep Reinforcement Learning Approach
Diego Patiño, Siddharth Mayya, Juan Calderon, Kostas Daniilidis, David Saldaña
Multi-Agent Reinforcement Learning for Cooperative Air Transportation Services in City-Wide Autonomous Urban Air Mobility
Chanyoung Park, Gyu Seon Kim, Soohyun Park, Soyi Jung, Joongheon Kim
Energy Management of Multi-mode Plug-in Hybrid Electric Vehicle using Multi-agent Deep Reinforcement Learning
Min Hua, Cetengfei Zhang, Fanggang Zhang, Zhi Li, Xiaoli Yu, Hongming Xu, Quan Zhou
Conditionally Optimistic Exploration for Cooperative Deep Multi-Agent Reinforcement Learning
Xutong Zhao, Yangchen Pan, Chenjun Xiao, Sarath Chandar, Janarthanan Rajendran