Motion Forecasting with Unlikelihood Training in Continuous Space

Deyao Zhu, Mohamed Zahran, Li Erran Li, Mohamed Elhoseiny
Proceedings of the 5th Conference on Robot Learning, PMLR 164:1003-1012, 2022.

Abstract

Motion forecasting is essential for making safe and intelligent decisions in robotic applications such as autonomous driving. Existing methods often formulate it as a sequence-to-sequence prediction problem, solved in an encoder-decoder framework with a maximum likelihood estimation objective. State-of-the-art models leverage contextual information including the map and states of surrounding agents. However, we observe that they still assign a high probability to unlikely trajectories resulting in unsafe behaviors including road boundary violations. Orthogonally, we propose a new objective, unlikelihood training, which forces predicted trajectories that conflict with contextual information to be assigned a lower probability. We demonstrate that our method can improve state-of-art models’ performance on challenging real-world trajectory forecasting datasets (nuScenes and Argoverse) by avoiding up to 56% context-violated prediction and improving up to 9% prediction accuracy. Code is avaliable at https://github.com/Vision-CAIR/UnlikelihoodMotionForecasting

Cite this Paper

BibTeX
@InProceedings{pmlr-v164-zhu22a, title = {Motion Forecasting with Unlikelihood Training in Continuous Space}, author = {Zhu, Deyao and Zahran, Mohamed and Li, Li Erran and Elhoseiny, Mohamed}, booktitle = {Proceedings of the 5th Conference on Robot Learning}, pages = {1003--1012}, year = {2022}, editor = {Faust, Aleksandra and Hsu, David and Neumann, Gerhard}, volume = {164}, series = {Proceedings of Machine Learning Research}, month = {08--11 Nov}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v164/zhu22a/zhu22a.pdf}, url = {https://proceedings.mlr.press/v164/zhu22a.html}, abstract = {Motion forecasting is essential for making safe and intelligent decisions in robotic applications such as autonomous driving. Existing methods often formulate it as a sequence-to-sequence prediction problem, solved in an encoder-decoder framework with a maximum likelihood estimation objective. State-of-the-art models leverage contextual information including the map and states of surrounding agents. However, we observe that they still assign a high probability to unlikely trajectories resulting in unsafe behaviors including road boundary violations. Orthogonally, we propose a new objective, unlikelihood training, which forces predicted trajectories that conflict with contextual information to be assigned a lower probability. We demonstrate that our method can improve state-of-art models’ performance on challenging real-world trajectory forecasting datasets (nuScenes and Argoverse) by avoiding up to 56% context-violated prediction and improving up to 9% prediction accuracy. Code is avaliable at https://github.com/Vision-CAIR/UnlikelihoodMotionForecasting} }
Endnote
%0 Conference Paper %T Motion Forecasting with Unlikelihood Training in Continuous Space %A Deyao Zhu %A Mohamed Zahran %A Li Erran Li %A Mohamed Elhoseiny %B Proceedings of the 5th Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2022 %E Aleksandra Faust %E David Hsu %E Gerhard Neumann %F pmlr-v164-zhu22a %I PMLR %P 1003--1012 %U https://proceedings.mlr.press/v164/zhu22a.html %V 164 %X Motion forecasting is essential for making safe and intelligent decisions in robotic applications such as autonomous driving. Existing methods often formulate it as a sequence-to-sequence prediction problem, solved in an encoder-decoder framework with a maximum likelihood estimation objective. State-of-the-art models leverage contextual information including the map and states of surrounding agents. However, we observe that they still assign a high probability to unlikely trajectories resulting in unsafe behaviors including road boundary violations. Orthogonally, we propose a new objective, unlikelihood training, which forces predicted trajectories that conflict with contextual information to be assigned a lower probability. We demonstrate that our method can improve state-of-art models’ performance on challenging real-world trajectory forecasting datasets (nuScenes and Argoverse) by avoiding up to 56% context-violated prediction and improving up to 9% prediction accuracy. Code is avaliable at https://github.com/Vision-CAIR/UnlikelihoodMotionForecasting
APA
Zhu, D., Zahran, M., Li, L.E. & Elhoseiny, M.. (2022). Motion Forecasting with Unlikelihood Training in Continuous Space. Proceedings of the 5th Conference on Robot Learning, in Proceedings of Machine Learning Research 164:1003-1012 Available from https://proceedings.mlr.press/v164/zhu22a.html.

Related Material