Goal Driven Multi-Robot Navigation in Simulated Environments with Federated Deep Reinforcement Learning

摘要

This research introduces a Federated Deep Reinforcement Learning (FDRL) approach utilizing the Twin Delayed Deep Deterministic Policy Gradients (TD3) algorithm to address challenges in Multi-Robot systems (MRSs). The main objective of this paper is to achieve autonomous movement of robots towards their designated destinations, incorporating obstacle avoidance and optimizing route selection in dynamic 3D environments. This is accomplished through the utilization of the federated-TD3 model. The proposed FDRL method is compared with the Federated Deep Deterministic Policy Gradients (Federated DDPG) algorithm and the traditional Multi Agent Deep Deterministic Policy Gradients (MADDPG), revealing superior stability and improved performance metrics, including goal completion percentage, average distance traveled, collision count, and average reward return. This study contributes to the advancement of efficient and reliable multi-robot systems, showcasing the efficacy of Federated TD3 in enhancing autonomy and navigation in simulated environments.