Trajectory-Level Data Augmentation for Offline Reinforcement Learning
Fuente:
arXiv
Salvato in:
| Autori principali: | , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910216138784768 |
|---|---|
| author | Schmähling, Tobias Burkhardt, Matthias Windisch, Tobias |
| author_facet | Schmähling, Tobias Burkhardt, Matthias Windisch, Tobias |
| contents | We propose a data augmentation method for offline reinforcement learning, motivated by active positioning problems. Particularly, our approach enables the training of off-policy models from a limited number of suboptimal trajectories. We introduce a trajectory-based augmentation technique that exploits task structure and the geometric relationship between rewards, value functions, and mathematical properties of logging policies. During data collection, our augmentation supports suboptimal logging policies, leading to higher data quality and improved offline reinforcement learning performance. We provide theoretical justification for these strategies and validate them empirically across positioning tasks of varying dimensionality and under partial observability. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_13401 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Trajectory-Level Data Augmentation for Offline Reinforcement Learning Schmähling, Tobias Burkhardt, Matthias Windisch, Tobias Machine Learning Robotics We propose a data augmentation method for offline reinforcement learning, motivated by active positioning problems. Particularly, our approach enables the training of off-policy models from a limited number of suboptimal trajectories. We introduce a trajectory-based augmentation technique that exploits task structure and the geometric relationship between rewards, value functions, and mathematical properties of logging policies. During data collection, our augmentation supports suboptimal logging policies, leading to higher data quality and improved offline reinforcement learning performance. We provide theoretical justification for these strategies and validate them empirically across positioning tasks of varying dimensionality and under partial observability. |
| title | Trajectory-Level Data Augmentation for Offline Reinforcement Learning |
| topic | Machine Learning Robotics |
| url | https://arxiv.org/abs/2605.13401 |