Decentralized secure multi-agent path planning using federated reinforcement learning and blockchain

Multi-agent path planning in decentralized settings presents issues such as limited communication, security risk, and scalability issues. Centralized approaches have a single point of failure and are not ideal to depend on. Our proposed Decentralized and Secure Multi-Agent Path Planning framework is based on Federated Reinforcement Learning (FRL) with Proximal Policy Optimization (PPO) and blockchain. This FRL-PPO framework allows agents to learn how to navigate effectively without transmitting raw data or unnecessary information, protecting agent privacy. Smart contracts based on blockchain technologies also facilitate secure communication and guarantee trust among agents. We demonstrated the value of the FRL-PPO configuration through experiments in a simulated environment that showed the speed of the learning process was enhanced, attack resistance, and the overall speed of path planning and path efficiency improved. Our approach reduces the risk of data manipulation, making autonomous multi-agent systems more secure, scalable, and effective in decentralized environments.