Predicting future trajectories of surrounding agents is essential for safety-critical autonomous driving. Most existing work focuses on predicting marginal trajectories for each agent independently. However, it has rarely been explored in predicting joint trajectories for →