| This paper presents a comparative analysis between two distinct deep reinforcement learning approaches for multi-agent coordination: the Deep Reinforcement Learning Market-Driven HierarchicalNeuro-Fuzzy Politree (DRL-MD-HNFP) and the Asynchronous Advantage Actor-Critic (A3C). While DRL-MD-HNFP utilizes explicit coordination mechanisms (Market-Driven and Coordination Graphs) within a hierarchical neuro-fuzzy architecture, A3C relies on a decentralized actor-critic approach with asynchronous parallel training. We evaluate both methods on two standard multi-agent reinforcement learning bench- marks: the pursuit-evasion game and a 3v3 robotic soccer simulation. Our experiments focus on comparing convergence speed, sample efficiency, final policy quality, and scalability. The results provide insights into the trade-offs between explicit, structured coordination and emergent coordination from decentralized learning, offering guidance for practitioners on selecting appropriate algorithms for different multi-agent problem domains |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.