The data used in validating autonomous vehicles is generally limited to real world datasets which cover just few traffic patterns. For example, only about 30 percent of the trajectories found in the Waymo Motion Dataset have durations exceeding 10 seconds. Thus, this thesis aims to use deep generative models to synthesize vehicle trajectories as they pass through a roundabout, given their entry and exit points. In this thesis two generative models were implemented and tested on several SUMO simulated vehicle trajectories in a Roundabout in Milan. A Generative Adversarial Network (GAN), and an autoregressive Transformer based model that is able to make multimodal predictions and generate motion.