Unseen vehicle trajectory prediction for safe autonomous driving

Abstract

In the task of autonomous driving, safe vehicle motion prediction is a challenging task not only due to the complicated street environment but also due to uncertain factors such as unseen vehicles that are outside the scope of the ego vehicle’s perception range. Henceforth, being able to predict potential trajectories of undetected vehicles is a necessary part of safe motion prediction. Traditional models rely on historical observation to predict the motion of surrounding vehicles, but what if there is no historical input? In this paper, we propose a method that can achieve motion prediction for both seen vehicles and unseen vehicles, taking both historical trajectories and statistical information as input and utilizing a self-devised deep-learning model. The prediction for unseen vehicle trajectories is represented in heatmaps. Experimental results show that our model is able to predict not only unseen vehicle trajectories that are in the dataset but also other potential ones that may appear in the real-world environment. To our best knowledge, safe motion prediction is a task that hasn’t been well explored, and we hope our work could inspire more to realize its importance.

Date of Award	2023
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology
Supervisor	Qifeng CHEN (Supervisor) & Lujia WANG (Supervisor)