Considerations About Using Large Language Models to Enhance the Prediction of Pedestrian Trajectories

Large Language Models (LLMs) mark the best era of artificial intelligence. These models are utilized across various industries and exhibit remarkable capabilities. We aim to investigate whether LLMs can enhance autonomous driving, focusing particularly on predicting pedestrian trajectories. This paper begins with an introduction to autonomous driving and a definition of the pedestrian trajectory prediction. The LLM components and approaches are discussed, along with their advantages, which can mitigate the current drawbacks in predicting pedestrian trajectories. The capabilities of LLMs in addressing this issue are demonstrated through recent architectures, such as LG-Traj and LMTraj. The downsides of using LLMs in autonomous driving are also summarized. Our work explores the integration of these Large Language Models into existing architectures of pedestrian trajectory prediction.