Embodiment-Classifier Guidance: Transferring Frozen VLAs Across Robot Dynamics without Demonstrations

Key points

• Flow-based policies such as $π0$ are trained on a fixed set of embodiments, and transferring them to new ones is an open problem.1
• On transfer tasks that $π0$ saw on its source body but the classifier was never trained on, we nudge each denoising step of $π0$ with the classifier's gradient, which estimates the direction from the distribution of successful actions on the source body toward that on the target body.1
• On the LIBERO simulation benchmark with four modified controllers, embodiment-classifier guidance raises $π0$'s average success on the target bodies from 72% to 80% across 12 suite-controller pairs, with a 31% relative reduction in failures (11 of 12 gain; 5 significant after Holm correction).1
• Collecting more target-body data with guided $π0$ can fill such gaps: on one weak-wrist arm, guided $π0$ solved a task on which unguided $π0$ had zero successes in the original data, and retraining the classifier on the guided successes raised success from 52.5% to 70.7%.1

Sources (1)

• [1]Embodiment-Classifier Guidance: Transferring Frozen VLAs Across Robot Dynamics without Demonstrations
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 15, 07:25 PM

  To adapt a frozen $π_0$ to a new embodiment without additional demonstrations or parameter updates, we train a lightweight embodiment classifier to distinguish successful action chunks on the source body from those on the target body, using a few calibration tasks.

  Flow-based policies such as $π_0$ are trained on a fixed set of embodiments, and transferring them to new ones is an open problem.

Extractive summary: sentences quoted from the sources.