Leveraging Semantic and Positional Uncertainty for Trajectory Prediction
Keywords: Uncertainty, Trajectory Prediction
TL;DR: We propose semantic and positional uncertainty to solve the problem of map noise in vehicle trajectory prediction so as to improve the performance of vehicle trajectory prediction
Abstract: Given a time horizon with historical movement data and environmental context, trajectory prediction aims to forecast the future motion of dynamic entities, such as vehicles and pedestrians. A key challenge in this task arises from the dynamic and noisy nature of real-time maps. This noise primarily stems from two resources: (1) positional errors due to sensor inaccuracies or environmental occlusions, and (2) cognitive errors resulting from incorrect scene understanding.
In an attempt to solve this problem, we propose a new framework that estimates two kinds of uncertainty, \ie, positional uncertainty and semantic uncertainty simultaneously, and explicitly incorporates both uncertainties into the trajectory prediction process.
In particular, we introduce a dual-head structure to independently perform semantic prediction twice and positional prediction twice, and further extract the prediction variance as the uncertainty indicator in an end-to-end manner. The uncertainty is then directly concatenated with the semantic and positional predictions to enhance the trajectory estimation.
To validate the effectiveness of our uncertainty-aware approach, we evaluate it on the real-world driving dataset, \ie, nuScenes.
Extensive experiments on 3 mapping estimation and 2 trajectory approaches show that the proposed method (1) effectively captures map noise through both positional and semantic uncertainties, and (2) seamlessly integrates and enhances existing trajectory prediction methods on multiple evaluation metrics, \ie, minADE, minFDE, and MR.
Primary Area: applications to computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 1508
Loading