Memorize, Adapt, Ignore: Diagnosing Robot Learning Mechanisms under Training Data Variation

What happened

arXiv:2609.38401v1 Announce Type: new Abstract: Training data variation, whether through designing a domain randomization (DR) scheme in simulation or curating demonstrations for imitation learning, is a primary lever for improving the robustness of robotic manipulation policies. Yet its underlying mechanisms remain poorly understood, and practitioners typically select randomization parameters through expensive trial and error.

An NTK-based signal-to-noise ratio also helps distinguish when policies have learned to \emph{ignore} task-irrelevant factors (e.g.\ treating blue and red cubes identically, instead of learning a blue sub-policy and a red sub-policy). Computer Science > Robotics arXiv:2609.38401 (cs) [Submitted on 29 Sep 2026] Title: Memorize, Adapt, Ignore: Diagnosing Robot Learning Mechanisms under Training Data Variation Authors: Ke Zhang , Danica J. arXiv is committed to these values and only works with partners that adhere to them.