Blockchain-assisted Distributed Dueling DQN for Multi-Agent Reinforcement Learning in Grid-Maze Environments

Abstract

Swarm robotics systems draw inspiration from the collective behavior of living organisms in nature, exhibiting flexible cooperation without a clear leader. Researchers have developed group robot systems that communicate with each other to achieve cooperative behavior and decision-making through distributed algorithms. In this study, we present a system that accomplishes shared goals through flexible task allocation among group robots using machine learning. Specifically, we develop and evaluate the effectiveness of multi-agent distributed cooperative operation algorithms using deep reinforcement learning. To enhance security, we encrypt the information shared by each robot using blockchain technology, which also reduces the computational load by dividing roles among robots. Our goal is to construct a cooperative and secure algorithm for performing work tasks as a group.