QSOD: Hybrid Policy Gradient for Deep Multi-agent Reinforcement Learning

When individuals interact with one another to accomplish specific goals, they learn from others’ experiences achieve the tasks at hand. The same holds for learning in virtual environments, such as video games. Deep multiagent reinforcement shows promising results terms of completing many challenging tasks. To demonstrate its viability, most algorithms use value decomposition multiple agents. guide each agent, behavior is utilized decompose combined Q-value agents into individual agent Q-values. A different mixing method can be utilized, using a monotonicity assumption based on QMIX and QVMix. However, this selects actions through greedy policy. agents, which require large numbers training trials, are not addressed. In paper, we propose novel hybrid policy action selection an known Selection Optimization DRL (QSOD). grey wolf optimizer (GWO) used determine choice individuals’ actions. As GWO, there proper attention among facilitated agents’ coordination another. We StarCraft 2 Learning Environment compare our proposed algorithm state-of-the-art Experimental that outperforms QVMix all scenarios requires fewer trials.