A Mechanistic Analysis of Sim-and-Real Co-Training in Generative Robot Policies
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_ | 1866911595216502784 |
|---|---|
| author | Lei, Yu Liu, Minghuan Maddukuri, Abhiram Jiang, Zhenyu Zhu, Yuke |
| author_facet | Lei, Yu Liu, Minghuan Maddukuri, Abhiram Jiang, Zhenyu Zhu, Yuke |
| contents | Co-training, which combines limited in-domain real-world data with abundant surrogate data such as simulation or cross-embodiment robot data, is widely used for training generative robot policies. Despite its empirical success, the mechanisms that determine when and why co-training is effective remain poorly understood. We investigate the mechanism of sim-and-real co-training through theoretical analysis and empirical study, and identify two intrinsic effects governing performance. The first, \textbf{``structured representation alignment"}, reflects a balance between cross-domain representation alignment and domain discernibility, and plays a primary role in downstream performance. The second, the \textbf{``importance reweighting effect"}, arises from domain-dependent modulation of action weighting and operates at a secondary level. We validate these effects with controlled experiments on a toy model and extensive sim-and-sim and sim-and-real robot manipulation experiments. Our analysis offers a unified interpretation of recent co-training techniques and motivates a simple method that consistently improves upon prior approaches. More broadly, our aim is to examine the inner workings of co-training and to facilitate research in this direction. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_13645 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | A Mechanistic Analysis of Sim-and-Real Co-Training in Generative Robot Policies Lei, Yu Liu, Minghuan Maddukuri, Abhiram Jiang, Zhenyu Zhu, Yuke Robotics Artificial Intelligence Machine Learning Co-training, which combines limited in-domain real-world data with abundant surrogate data such as simulation or cross-embodiment robot data, is widely used for training generative robot policies. Despite its empirical success, the mechanisms that determine when and why co-training is effective remain poorly understood. We investigate the mechanism of sim-and-real co-training through theoretical analysis and empirical study, and identify two intrinsic effects governing performance. The first, \textbf{``structured representation alignment"}, reflects a balance between cross-domain representation alignment and domain discernibility, and plays a primary role in downstream performance. The second, the \textbf{``importance reweighting effect"}, arises from domain-dependent modulation of action weighting and operates at a secondary level. We validate these effects with controlled experiments on a toy model and extensive sim-and-sim and sim-and-real robot manipulation experiments. Our analysis offers a unified interpretation of recent co-training techniques and motivates a simple method that consistently improves upon prior approaches. More broadly, our aim is to examine the inner workings of co-training and to facilitate research in this direction. |
| title | A Mechanistic Analysis of Sim-and-Real Co-Training in Generative Robot Policies |
| topic | Robotics Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2604.13645 |