EscIRL: Evolving Self-Contrastive IRL for Trajectory Prediction in Autonomous Driving

Siyue Wang, Zhaorun Chen, Zhuokai Zhao, Chaoli Mao, Yiyang Zhou, Jiayu He, Albert Sibo Hu
Proceedings of The 8th Conference on Robot Learning, PMLR 270:5446-5465, 2025.

Abstract

While deep neural networks (DNN) and inverse reinforcement learning (IRL) have both been commonly used in autonomous driving to predict trajectories through learning from expert demonstrations, DNN-based methods suffer from data-scarcity, while IRL-based approaches often struggle with generalizability, making both hard to apply to new driving scenarios. To address these issues, we introduce EscIRL, a novel decoupled bi-level training framework that iteratively learns robust reward models from only a few mixed-scenario demonstrations. At the inner level, EscIRL introduces a self-contrastive IRL module that learns a spectrum of specialized reward functions by contrasting demonstrations across different scenarios. At the outer level, ESCIRL employs an evolving loop that iteratively refines the contrastive sets, ensuring global convergence. Experiments on two multi-scenario datasets, CitySim and INTERACTION, demonstrate the effectiveness of EscIRL, outperforming state-of-the-art DNN and IRL-based methods by 41.3% on average. Notably, we show that EscIRL achieves superior generalizability compared to DNN-based approaches while requiring only a small fraction of the data, effectively addressing data-scarcity constraints. All code and data are available at https://github.com/SiyueWang-CiDi/EscIRL.

Cite this Paper

BibTeX
@InProceedings{pmlr-v270-wang25n, title = {EscIRL: Evolving Self-Contrastive IRL for Trajectory Prediction in Autonomous Driving}, author = {Wang, Siyue and Chen, Zhaorun and Zhao, Zhuokai and Mao, Chaoli and Zhou, Yiyang and He, Jiayu and Hu, Albert Sibo}, booktitle = {Proceedings of The 8th Conference on Robot Learning}, pages = {5446--5465}, year = {2025}, editor = {Agrawal, Pulkit and Kroemer, Oliver and Burgard, Wolfram}, volume = {270}, series = {Proceedings of Machine Learning Research}, month = {06--09 Nov}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v270/main/assets/wang25n/wang25n.pdf}, url = {https://proceedings.mlr.press/v270/wang25n.html}, abstract = {While deep neural networks (DNN) and inverse reinforcement learning (IRL) have both been commonly used in autonomous driving to predict trajectories through learning from expert demonstrations, DNN-based methods suffer from data-scarcity, while IRL-based approaches often struggle with generalizability, making both hard to apply to new driving scenarios. To address these issues, we introduce EscIRL, a novel decoupled bi-level training framework that iteratively learns robust reward models from only a few mixed-scenario demonstrations. At the inner level, EscIRL introduces a self-contrastive IRL module that learns a spectrum of specialized reward functions by contrasting demonstrations across different scenarios. At the outer level, ESCIRL employs an evolving loop that iteratively refines the contrastive sets, ensuring global convergence. Experiments on two multi-scenario datasets, CitySim and INTERACTION, demonstrate the effectiveness of EscIRL, outperforming state-of-the-art DNN and IRL-based methods by 41.3% on average. Notably, we show that EscIRL achieves superior generalizability compared to DNN-based approaches while requiring only a small fraction of the data, effectively addressing data-scarcity constraints. All code and data are available at https://github.com/SiyueWang-CiDi/EscIRL.} }
Endnote
%0 Conference Paper %T EscIRL: Evolving Self-Contrastive IRL for Trajectory Prediction in Autonomous Driving %A Siyue Wang %A Zhaorun Chen %A Zhuokai Zhao %A Chaoli Mao %A Yiyang Zhou %A Jiayu He %A Albert Sibo Hu %B Proceedings of The 8th Conference on Robot Learning %C Proceedings of Machine Learning Research %D 2025 %E Pulkit Agrawal %E Oliver Kroemer %E Wolfram Burgard %F pmlr-v270-wang25n %I PMLR %P 5446--5465 %U https://proceedings.mlr.press/v270/wang25n.html %V 270 %X While deep neural networks (DNN) and inverse reinforcement learning (IRL) have both been commonly used in autonomous driving to predict trajectories through learning from expert demonstrations, DNN-based methods suffer from data-scarcity, while IRL-based approaches often struggle with generalizability, making both hard to apply to new driving scenarios. To address these issues, we introduce EscIRL, a novel decoupled bi-level training framework that iteratively learns robust reward models from only a few mixed-scenario demonstrations. At the inner level, EscIRL introduces a self-contrastive IRL module that learns a spectrum of specialized reward functions by contrasting demonstrations across different scenarios. At the outer level, ESCIRL employs an evolving loop that iteratively refines the contrastive sets, ensuring global convergence. Experiments on two multi-scenario datasets, CitySim and INTERACTION, demonstrate the effectiveness of EscIRL, outperforming state-of-the-art DNN and IRL-based methods by 41.3% on average. Notably, we show that EscIRL achieves superior generalizability compared to DNN-based approaches while requiring only a small fraction of the data, effectively addressing data-scarcity constraints. All code and data are available at https://github.com/SiyueWang-CiDi/EscIRL.
APA
Wang, S., Chen, Z., Zhao, Z., Mao, C., Zhou, Y., He, J. & Hu, A.S.. (2025). EscIRL: Evolving Self-Contrastive IRL for Trajectory Prediction in Autonomous Driving. Proceedings of The 8th Conference on Robot Learning, in Proceedings of Machine Learning Research 270:5446-5465 Available from https://proceedings.mlr.press/v270/wang25n.html.

Related Material