Federated RL

Federated reinforcement learning studies how agents can improve decision-making together by sharing model updates rather than raw trajectories. Common approaches adapt federated averaging, actor-critic and Q-learning methods, and add personalization or meta-learning when agents face different environments; decentralized aggregation, robust statistics, and compressed or event-triggered communication address other practical constraints. Applications include robotics and UAV teams, wireless and edge resource allocation, autonomous vehicles, and energy systems, often evaluated in control benchmarks or network simulations. A central challenge is balancing shared learning with local adaptation while handling non-IID data, limited communication, unreliable participants, and safety constraints; theoretical guarantees are developing, but real-world deployment remains difficult to establish.