Learning Self-Correction in Vision–Language Models via Rollout Augmentation

Yi Ding, Ziliang Qiu, Bolian Li, Ruqi Zhang
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:25096-25113, 2026.

Abstract

Self-correction is essential for solving complex reasoning problems in vision–language models (VLMs), yet existing reinforcement learning (RL) methods struggle to learn it. Effective self-correction behaviors emerge only rarely during RL, making learning signals sparse. To address this challenge, we propose correction-specific rollouts (Octopus), a rollout-augmentation framework that synthesizes dense self-correction supervision by recombining existing rollouts without computational overhead. This rollout augmentation simultaneously improves sample efficiency and stabilizes RL optimization. Furthermore, we introduce a two-stage RL training strategy that disentangles self-correction and direct reasoning, avoiding signal conflicts and enabling both behaviors to be learned effectively. Building on this, we introduce $\texttt{Octopus-8B}$, an advanced reasoning VLM with controllable self-correction capabilities. It achieves SoTA performance among open-source VLMs across 7 benchmarks, outperforming the best RLVR baseline by 1.0 score while requiring only $0.72\times$ training time per step.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-ding26m, title = {Learning Self-Correction in Vision–Language Models via Rollout Augmentation}, author = {Ding, Yi and Qiu, Ziliang and Li, Bolian and Zhang, Ruqi}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {25096--25113}, 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/ding26m/ding26m.pdf}, url = {https://proceedings.mlr.press/v306/ding26m.html}, abstract = {Self-correction is essential for solving complex reasoning problems in vision–language models (VLMs), yet existing reinforcement learning (RL) methods struggle to learn it. Effective self-correction behaviors emerge only rarely during RL, making learning signals sparse. To address this challenge, we propose correction-specific rollouts (Octopus), a rollout-augmentation framework that synthesizes dense self-correction supervision by recombining existing rollouts without computational overhead. This rollout augmentation simultaneously improves sample efficiency and stabilizes RL optimization. Furthermore, we introduce a two-stage RL training strategy that disentangles self-correction and direct reasoning, avoiding signal conflicts and enabling both behaviors to be learned effectively. Building on this, we introduce $\texttt{Octopus-8B}$, an advanced reasoning VLM with controllable self-correction capabilities. It achieves SoTA performance among open-source VLMs across 7 benchmarks, outperforming the best RLVR baseline by 1.0 score while requiring only $0.72\times$ training time per step.} }
Endnote
%0 Conference Paper %T Learning Self-Correction in Vision–Language Models via Rollout Augmentation %A Yi Ding %A Ziliang Qiu %A Bolian Li %A Ruqi Zhang %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-ding26m %I PMLR %P 25096--25113 %U https://proceedings.mlr.press/v306/ding26m.html %V 306 %X Self-correction is essential for solving complex reasoning problems in vision–language models (VLMs), yet existing reinforcement learning (RL) methods struggle to learn it. Effective self-correction behaviors emerge only rarely during RL, making learning signals sparse. To address this challenge, we propose correction-specific rollouts (Octopus), a rollout-augmentation framework that synthesizes dense self-correction supervision by recombining existing rollouts without computational overhead. This rollout augmentation simultaneously improves sample efficiency and stabilizes RL optimization. Furthermore, we introduce a two-stage RL training strategy that disentangles self-correction and direct reasoning, avoiding signal conflicts and enabling both behaviors to be learned effectively. Building on this, we introduce $\texttt{Octopus-8B}$, an advanced reasoning VLM with controllable self-correction capabilities. It achieves SoTA performance among open-source VLMs across 7 benchmarks, outperforming the best RLVR baseline by 1.0 score while requiring only $0.72\times$ training time per step.
APA
Ding, Y., Qiu, Z., Li, B. & Zhang, R.. (2026). Learning Self-Correction in Vision–Language Models via Rollout Augmentation. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:25096-25113 Available from https://proceedings.mlr.press/v306/ding26m.html.

Related Material