The technology of dynamic map fusion among networked vehicles has beendeveloped to enlarge sensing ranges and improve sensing accuracies forindividual vehicles. This paper proposes a federated learning (FL) baseddynamic map fusion framework to achieve high map quality despite unknownnumbers of objects in fields of view (FoVs), various sensing and modeluncertainties, and missing data labels for online learning. The novelty of thiswork is threefold: (1) developing a three-stage fusion scheme to predict thenumber of objects effectively and to fuse multiple local maps with fidelityscores; (2) developing an FL algorithm which fine-tunes feature models (i.e.,representation learning networks for feature extraction) distributively byaggregating model parameters; (3) developing a knowledge distillation method togenerate FL training labels when data labels are unavailable. The proposedframework is implemented in the Car Learning to Act (CARLA) simulationplatform. Extensive experimental results are provided to verify the superiorperformance and robustness of the developed map fusion and FL schemes.