[2507.13340] Latent Policy Steering with Embodiment-Agnostic Pretrained World Models

Computer Science > Robotics arXiv:2507.13340 (cs) [Submitted on 17 Jul 2025 (v1), last revised 22 Mar 2026 (this version, v4)] Title:Latent Policy Steering with Embodiment-Agnostic Pretrained World Models Authors:Yiqi Wang, Mrinal Verghese, Jeff Schneider View a PDF of the paper titled Latent Policy Steering with Embodiment-Agnostic Pretrained World Models, by Yiqi Wang and 1 other authors View PDF HTML (experimental) Abstract:The performance of learned robot visuomotor policies is heavily dependent on the size and quality of the training dataset. Although large-scale robot and human datasets are increasingly available, embodiment gaps and mismatched action spaces make them difficult to leverage. Our main insight is that skills performed across different embodiments produce visual similarities in motions that can be captured using off-the-shelf action representations such as optical flow. Moreover, World Models (WMs) can leverage sub-optimal data since they focus on modeling dynamics. In this work, we aim to improve visuomotor policies in low-data regimes by first pretraining a WM using optical flow as an embodiment-agnostic action representation to leverage accessible or easily collected data from multiple embodiments (robots, humans). Given a small set of demonstrations on a target embodiment, we finetune the WM on this data to better align the WM predictions, train a base policy, and learn a robust value function. Using our finetuned WM and value function, our approach ...