Accurate prediction of road user behavior is a fundamental capability for autonomous vehicles, enabling safe and informed decision-making. However, manually modeling the intricate interactions between agents and their environment is challenging due to the complexity and variability of real-world driving scenarios. Traditional rule-based and analytical approaches struggle to generalize across diverse traffic situations, necessitating data-driven methods that automatically learn these patterns from large-scale datasets. Deep learning has emerged as a key enabler for capturing these high-dimensional dependencies.
This project explores deep learning-based trajectory prediction, focusing on handling the multimodality and uncertainty inherent in traffic behaviors. To this end, we investigate generative and discriminative approaches, such as Diffusion and Mixture Density Networks (MDNs) for parameterizing probabilistic multimodal predictions. We aim to advance scalable, probabilistic trajectory forecasting models that enhance the robustness and reliability of autonomous decision-making.