Definition
Plain language
A training leash that lets a robot explore freely while constantly tugging it back toward imitating the human demonstrations.
As stated in the literature
An RL technique that adds a term penalizing deviation from demonstrated behavior, keeping exploration anchored to a sensible baseline; in ENPIRE it produced the single largest jump in pin-insertion success.
Also called: BC regularization
Why it matters: It lets a robot explore for better strategies without wandering into useless behavior, and here it drove the biggest single gain in task success.
For example, as a robot tries new ways to insert a pin, this technique keeps nudging it back toward the steady motions a human showed it.
Heard on the show
“The standout: a trick called behavior-cloning regularization jumped success by almost eleven percentage points in a single step.”Episode 159 — Can a Coding Agent Run Its Own Robot Experiments Overnight, With No Human Resetting the Scene?