Trajectory prediction is essential for ensuring safety in autonomous driving by anticipating the movements of traffic participants such as vehicles and pedestrians. While traditional deep learning approaches have enhanced prediction accuracy, they face significant shortcomings including interpretability issues and dependence on extensive datasets. The emergence of Large Foundation Models (LFMs), particularly Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs), has introduced a transformative shift in trajectory prediction methodologies, enhancing the capability to reason contextually.
This survey outlines key methodologies like trajectory-language mapping, multimodal fusion, and constraint-based reasoning which contribute to more robust and interpretable predictions. It discusses crucial challenges faced in the field, such as computational latency and data scarcity, which hinder deployment in real-world scenarios. As the research landscape evolves, future directions focus on achieving low-latency inference and developing causality-aware models, indicating a promising trajectory for improving the safety and reliability of autonomous driving systems.
👉 Pročitaj original: arXiv AI Papers
