Transport or Discard: Robust Unbalanced Optimal Transport for Cross-Domain Policy Adaptation

Wenyu Chen, Yujia Zhang, Wei Guo, Linli Ma, Yanbo Wang, Pinle Qin, Jianchao Zeng
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:18099-18123, 2026.

Abstract

Cross-domain offline reinforcement learning leverages a source dataset to improve policy learning in a data-scarce target domain, but dynamics mismatch makes many source transitions kinematically infeasible and can cause negative transfer. Recent non-parametric geometric methods (e.g., standard optimal transport and k-nearest neighbors) avoid overfitting yet often yield only relative rankings under an implicit matching or retrieval budget, making performance sensitive to hand-tuned thresholds when the true cross-domain overlap is unknown. We formulate availability estimation as soft subset selection by learning a source reweighting that geometrically aligns with the target. We propose Robust Offline unbalanced Optimal Transport (ROOT): (i) a robust ambiguity set for uncertainty under limited target samples, and (ii) an unbalanced transport objective that penalizes mass deviation, enabling a principled transport-or-discard mechanism. ROOT thus down-weights or discards high-cost source samples rather than forcing them onto the target support. Moreover, the induced weights decay exponentially with transport cost, guaranteeing outlier suppression. On D4RL dynamics-shift benchmarks, ROOT improves downstream offline RL and outperforms strong baselines on most tasks without task-specific threshold tuning.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-chen26gd, title = {Transport or Discard: Robust Unbalanced Optimal Transport for Cross-Domain Policy Adaptation}, author = {Chen, Wenyu and Zhang, Yujia and Guo, Wei and Ma, Linli and Wang, Yanbo and Qin, Pinle and Zeng, Jianchao}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {18099--18123}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/chen26gd/chen26gd.pdf}, url = {https://proceedings.mlr.press/v306/chen26gd.html}, abstract = {Cross-domain offline reinforcement learning leverages a source dataset to improve policy learning in a data-scarce target domain, but dynamics mismatch makes many source transitions kinematically infeasible and can cause negative transfer. Recent non-parametric geometric methods (e.g., standard optimal transport and k-nearest neighbors) avoid overfitting yet often yield only relative rankings under an implicit matching or retrieval budget, making performance sensitive to hand-tuned thresholds when the true cross-domain overlap is unknown. We formulate availability estimation as soft subset selection by learning a source reweighting that geometrically aligns with the target. We propose Robust Offline unbalanced Optimal Transport (ROOT): (i) a robust ambiguity set for uncertainty under limited target samples, and (ii) an unbalanced transport objective that penalizes mass deviation, enabling a principled transport-or-discard mechanism. ROOT thus down-weights or discards high-cost source samples rather than forcing them onto the target support. Moreover, the induced weights decay exponentially with transport cost, guaranteeing outlier suppression. On D4RL dynamics-shift benchmarks, ROOT improves downstream offline RL and outperforms strong baselines on most tasks without task-specific threshold tuning.} }
Endnote
%0 Conference Paper %T Transport or Discard: Robust Unbalanced Optimal Transport for Cross-Domain Policy Adaptation %A Wenyu Chen %A Yujia Zhang %A Wei Guo %A Linli Ma %A Yanbo Wang %A Pinle Qin %A Jianchao Zeng %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-chen26gd %I PMLR %P 18099--18123 %U https://proceedings.mlr.press/v306/chen26gd.html %V 306 %X Cross-domain offline reinforcement learning leverages a source dataset to improve policy learning in a data-scarce target domain, but dynamics mismatch makes many source transitions kinematically infeasible and can cause negative transfer. Recent non-parametric geometric methods (e.g., standard optimal transport and k-nearest neighbors) avoid overfitting yet often yield only relative rankings under an implicit matching or retrieval budget, making performance sensitive to hand-tuned thresholds when the true cross-domain overlap is unknown. We formulate availability estimation as soft subset selection by learning a source reweighting that geometrically aligns with the target. We propose Robust Offline unbalanced Optimal Transport (ROOT): (i) a robust ambiguity set for uncertainty under limited target samples, and (ii) an unbalanced transport objective that penalizes mass deviation, enabling a principled transport-or-discard mechanism. ROOT thus down-weights or discards high-cost source samples rather than forcing them onto the target support. Moreover, the induced weights decay exponentially with transport cost, guaranteeing outlier suppression. On D4RL dynamics-shift benchmarks, ROOT improves downstream offline RL and outperforms strong baselines on most tasks without task-specific threshold tuning.
APA
Chen, W., Zhang, Y., Guo, W., Ma, L., Wang, Y., Qin, P. & Zeng, J.. (2026). Transport or Discard: Robust Unbalanced Optimal Transport for Cross-Domain Policy Adaptation. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:18099-18123 Available from https://proceedings.mlr.press/v306/chen26gd.html.

Related Material