A Federated Meta-Reinforcement Learning Algorithm Based on Gradient Correction

To meet the requirement of the decision-making capabilities of IoT devices in dynamic environments, Federated Reinforcement Learning (FRL) has attracted increasing attention in recent years. However, Due to the challenges of heterogeneous environments and sampling limitation in the practical scenarios, it is difficult for clients in different environments to jointly train a personalized policy that can quickly adapt to a new environment. In this paper, we propose a federated meta-reinforcement learning algorithm with gradient correction to extract a meta-policy from heterogeneous environments. Further, we analyze the performance of our proposed algorithm in the experiment section.