Learning in Cooperative Multiagent Systems Using Cognitive and Machine Models

Developing effective Multi-Agent Systems (MAS) is critical for many applications requiring collaboration and coordination with humans. Despite the rapid advance of Deep Reinforcement Learning (MADRL) in cooperative MAS, one major challenges that remain simultaneous learning interaction independent agents dynamic environments presence stochastic rewards. State-of-the-art MADRL models struggle to perform well Coordinated Multi-agent Object Transportation Problems (CMOTPs) wherein must coordinate each other learn from In contrast, humans often rapidly adapt nonstationary require among people. this paper, motivated by demonstrated ability cognitive based on Instance-Based Theory (IBLT) capture human decisions decision making tasks, we propose three variants IBL (MAIBL). The idea these MAIBL algorithms combine mechanisms IBLT techniques deal MAS perspective learners. We demonstrate exhibit faster achieve better a CMOTP task various settings rewards compared current models. discuss benefits integrating insights into