RePO-VLA: Recovery-Driven Policy Optimization for Vision-Language-Action Models
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915998761746432 |
|---|---|
| author | Liufu, Weijia Guo, Xiaoyu Chen, Ruiyi Liu, Jingzhi Zhang, Kaidong Liang, Xiwen Lin, Jianqi Sun, Dawei Wang, Yuze Xu, Rongtao Lin, Bingqian Yang, Bowen Cao, Tongtong Peng, Bowen Zhang, Dongyu Wang, Guangrun Wang, Min Lin, Liang Liang, Xiaodan |
| author_facet | Liufu, Weijia Guo, Xiaoyu Chen, Ruiyi Liu, Jingzhi Zhang, Kaidong Liang, Xiwen Lin, Jianqi Sun, Dawei Wang, Yuze Xu, Rongtao Lin, Bingqian Yang, Bowen Cao, Tongtong Peng, Bowen Zhang, Dongyu Wang, Guangrun Wang, Min Lin, Liang Liang, Xiaodan |
| contents | Vision-Language-Action (VLA) models remain brittle in long-horizon, contact-rich manipulation because success-only imitation provides little supervision for execution drift, while failed rollouts are often discarded. We introduce RePO-VLA, a recovery-driven policy optimization framework that assigns distinct roles to success, recovery, and failure trajectories. RePO-VLA first applies Recovery-Aware Initialization (RAI), slicing recovery segments and resetting history so corrective actions depend on the current adverse state rather than the preceding failure. It then learns a Progress-Aware Semantic Value Function (PAS-VF), aligning spatiotemporal trajectory features with instructions and successful references. The resulting labels salvage useful failure prefixes via reliability decay, while low-value labels mark drift and terminal breakdowns, teaching differences among nominal, failed, and corrective actions. The data engine turns adverse states into planner-generated or human-collected corrective rollouts, teaching recovery to the success manifold. Value-Conditioned Refinement (VCR) trains the policy to prefer high-progress actions. At deployment, a fixed high value ($v=1.0$) biases actions toward the learned success manifold without online failure detectors or heuristic retries. We introduce FRBench, with standardized error injection and recovery-focused evaluation. Across simulated and real-world bimanual tasks, RePO-VLA improves robustness, raising adversarial success from 20% to 75% on average and up to 80% in scaled real-world trials. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_09410 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | RePO-VLA: Recovery-Driven Policy Optimization for Vision-Language-Action Models Liufu, Weijia Guo, Xiaoyu Chen, Ruiyi Liu, Jingzhi Zhang, Kaidong Liang, Xiwen Lin, Jianqi Sun, Dawei Wang, Yuze Xu, Rongtao Lin, Bingqian Yang, Bowen Cao, Tongtong Peng, Bowen Zhang, Dongyu Wang, Guangrun Wang, Min Lin, Liang Liang, Xiaodan Robotics Artificial Intelligence Vision-Language-Action (VLA) models remain brittle in long-horizon, contact-rich manipulation because success-only imitation provides little supervision for execution drift, while failed rollouts are often discarded. We introduce RePO-VLA, a recovery-driven policy optimization framework that assigns distinct roles to success, recovery, and failure trajectories. RePO-VLA first applies Recovery-Aware Initialization (RAI), slicing recovery segments and resetting history so corrective actions depend on the current adverse state rather than the preceding failure. It then learns a Progress-Aware Semantic Value Function (PAS-VF), aligning spatiotemporal trajectory features with instructions and successful references. The resulting labels salvage useful failure prefixes via reliability decay, while low-value labels mark drift and terminal breakdowns, teaching differences among nominal, failed, and corrective actions. The data engine turns adverse states into planner-generated or human-collected corrective rollouts, teaching recovery to the success manifold. Value-Conditioned Refinement (VCR) trains the policy to prefer high-progress actions. At deployment, a fixed high value ($v=1.0$) biases actions toward the learned success manifold without online failure detectors or heuristic retries. We introduce FRBench, with standardized error injection and recovery-focused evaluation. Across simulated and real-world bimanual tasks, RePO-VLA improves robustness, raising adversarial success from 20% to 75% on average and up to 80% in scaled real-world trials. |
| title | RePO-VLA: Recovery-Driven Policy Optimization for Vision-Language-Action Models |
| topic | Robotics Artificial Intelligence |
| url | https://arxiv.org/abs/2605.09410 |