LSTR: Long-Short Range Aggregation for Trajectory Prediction at Intersection Scenarios
Keywords: motion prediction, autonomous driving, path_planning
TL;DR: We propose a novel Long-short Range aggregation for trajectory prediction at intersection scenarios
Abstract: Trajectory prediction is crucial for practical applications, encompassing navigation for autonomous vehicles and the implementation of safety systems based on the Internet of Vehicles (IoV). Most existing methods significantly rely on comprehensive map information, employing robust rule constraints to incrementally predict trajectories within the driver's local decision-making context. However, in environments characterized by weak rule enforcement, such as urban intersections, these approaches neglect the disparity between the driver's overarching intentions and current behaviors.Recognizing the characteristics of intersection traffic flow—macroscopically organized yet microscopically disordered, exhibiting highly heterogeneous conditions—this paper presents a novel model termed Long-short Range Aggregation for Trajectory Prediction in Intersections (LSTR). This model anchors the vehicle's local decision-making process to long-range intentions. Specifically, LSTR predicts the vehicle's destination via a global intention inference module and models its long-range driving intentions through clustering to extract macroscopic traffic flow patterns. This long-range intention subsequently informs the short-range local interaction behaviors captured by the local behavior decision module. Ultimately, the fused features from these two modules are analyzed using a multi-modal decoder to interpret the various motion patterns, resulting in the trajectory prediction outcomes.We rigorously validate the proposed framework across multiple intersection scenarios utilizing real-world datasets, including inD, roundD, and a subset of WOMD. Experimental results demonstrate that our model outperforms numerous benchmarks without relying on additional information such as HD maps of intersections.
Primary Area: applications to robotics, autonomy, planning
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: 7278
Loading