Latent policy steering improves robot learning across different bodies
Latent Policy Steering: An Efficient and Flexible Framework for Cross-Embodiment Transfer
Robotics
Summary
Teaching robots to do tasks well needs a lot of good training examples, which can be hard to get for each type of robot. The authors found a way to reuse lessons learned from different kinds of robots by focusing on how the world changes when any robot moves, instead of directly copying actions. They built a system called Latent Policy Steering that first learns general visual dynamics from many robots and then adapts quickly to a new robot with few examples. This method helps robots learn faster and do better with less data, even when their bodies are very different.
What this means in practice
- •For robotics engineers: Improve robot task learning efficiency by adapting policies learned from other robot types with limited new data.
- •For automation system integrators: Deploy versatile robot control across different machines with fewer demonstrations by using a shared visual dynamics model.
Authors
Yiqi Wang, Mrinal Verghese, Jeff Schneider
Abstract
The performance of learned robot visuomotor policies depends heavily on the size and quality of their training data, yet collecting high-quality demonstrations remains costly for robots in the real world. Although large-scale robot and human datasets are increasingly available, embodiment gaps and mismatched action spaces make them difficult to leverage directly. Cross-embodiment transfer, reusing experience from other embodiments to improve learning on a target embodiment, is therefore crucial for scaling robot learning beyond per-robot data collection. In this work, we find that efficient transfer can be achieved by learning from what is shared across embodiments, the visual dynamics of how the world responds to motion, and by effectively exploiting the scarce target-embodiment data at test time. The proposed framework, called Latent Policy Steering (LPS), implements an embodiment-agnostic pretraining phase, which trains an image-based World Model (WM) with optical flow across diverse embodiments. The resulting WM is finetuned on the target embodiment with robot actions. It then steers the base policy toward better actions by searching in the WM's latent space for plans that stay close to the finetuning data. LPS is a policy-agnostic framework: it can flexibly accommodate different policies without having to retrain them. In Robomimic and real-world evaluations, LPS improves the average performance of Diffusion Policy relatively by 16% and 62%, and Pi0.5 by 8% and 14%, with only 50 demonstrations on an unseen target embodiment.