A2CM: a new multi-agent algorithm

Abstract

Reinforcement learning is currently one of the most researched fields of artificial intelligence. New algorithms are being developed that use neural networks to compute the selected action, especially for deep reinforcement learning. One subcategory of reinforcement learning is multi-agent reinforcement learning, in which multiple agents are present in the world. As it involves the simulation of an environment, it can be applied to robotics as well. In our paper, we use our modified version of the advantage actor–critic (A2C) algorithm, which is suitable for multi-agent scenarios. We test this modified algorithm on our testbed, a cooperative–competitive pursuit–evasion environment, and later we address the problem of collision avoidance.