The federated learning models are currently employed in VANETs to improve the learning efficiency with minimum transmission overhead. The federated learning-based models preserve the privacy level of VANET users by only transmitting the model updates of the learnable parameters instead of transmitting the entire parameters of a global dataset. The major learning and communication challenges of implementing federated learning-based VANET models are data heterogeneity, labeling parameters, effective model training, security, transmission delay, computation cost, efficient scheduling, and resource management. The training procedure of federated learning models is similar to machine learning models, whereas they do not involve in entire dataset transmission. Thus, it minimizes the overhead considerably while preserving the privacy level of users.