federated-reinforcement-learning

This paper proposes FedQHD, a novel federated Q-learning method using hyperdimensional random-feature state encoders with linear readouts to enable closed-form function-space aggregation, addressing the federation gap due to heterogeneous client encoders.