Reaction-Uncertainty-Aware Motion Planning for Interactive Driving via Conditional Multi-Modal Motion Prediction
IEEE Access · 2026 · 14 · 51250-51262
DOI: 10.1109/access.2026.3679602 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Hanbin Jang und 3 weiteren Autor:innen. Die Arbeit nennt 40 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Ministry of Land, Infrastructure and Transport (MOLIT) of the Republic of Korea, Ministry of Trade, Industry and Energy (MOTIE) and Korea Evaluation Institute of Industrial Technology (KEIT) of the Republic of Korea.
Abstract (Originalsprache)
For autonomous vehicles to drive safely and in a socially collaborative manner, it is essential to consider interactions with surrounding agents. To model these interactions, recent motion planning works employ conditional motion predictions, which model the impact of the ego vehicle’s plan on surrounding agents by treating the ego’s future plan as query for prediction. However, most existing planning studies rely on uni-modal conditional motion prediction models, which struggle to represent the uncertain intentions of surrounding agents. Although a few studies adopt multi-modal mechanisms, they still fail to model reaction uncertainty—how surrounding agents respond to the ego’s future plan. In this paper, we propose a reaction-uncertainty-aware motion planning framework that integrates a newly designed conditional multi-modal motion prediction (CMMP) with a tree-structured policy planner (TPP). CMMP effectively models the reaction uncertainty of surrounding agents through multi-modal and causal masked attention mechanisms. TPP generates tree structured ego-trajectory candidates and probabilistically evaluates them based on the multi-modal prediction results to enhance the robustness of decision-making. The proposed framework is validated on the large-scale real-world nuPlan dataset. Experimental results demonstrate that, compared to the baseline relying on uni-modal conditional motion prediction, our method achieves up to a 32.39% improvement in planning performance, particularly excelling in interactive scenarios.
Förderung
- Ministry of Land, Infrastructure and Transport (MOLIT) of the Republic of Korea
- Ministry of Trade, Industry and Energy (MOTIE) and Korea Evaluation Institute of Industrial Technology (KEIT) of the Republic of Korea
Zitieren
APA
Jang, H., Lim, J., Syamil, A. R., & Kum, D. (2026). Reaction-Uncertainty-Aware Motion Planning for Interactive Driving via Conditional Multi-Modal Motion Prediction. IEEE Access, 14, 51250-51262. https://doi.org/10.1109/access.2026.3679602
Vancouver
Jang H, Lim J, Syamil AR, Kum D. Reaction-Uncertainty-Aware Motion Planning for Interactive Driving via Conditional Multi-Modal Motion Prediction. IEEE Access. 2026;14:51250-51262. doi:10.1109/access.2026.3679602
BibTeX
@article{Jang2026,
author = {Jang, Hanbin and Lim, Joonhee and Syamil, Abi Rahman and Kum, Dongsuk},
title = {Reaction-Uncertainty-Aware Motion Planning for Interactive Driving via Conditional Multi-Modal Motion Prediction},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {51250--51262},
doi = {10.1109/access.2026.3679602}
}
EUVIMED