Event
SCRIM: Quantifying State Criticality in Imitation Learning
1 reports · 1 sources
Yusuf Umut Ciftci and co-authors propose SCRIM, a state criticality measure for imitation learning: around a fixed imitation policy, they train an adversarial perturbation policy and a value critic to estimate the worst-case downstream value degradation caused by small perturbations to expert actions, assigning a score to each demonstrated state. The authors state that these scores can be used to reweight the behavior cloning loss, improving imitation at critical states and enhancing overall task performance; they can also be used to select a compact subset of high-value demonstrations, substantially reducing the amount of training data while maintaining performance. The authors evaluated the method on a variety of simulated and real robot manipulation tasks, and the paper was accepted to CoRL 2026. The above results are as reported by the authors.
Report timeline
- SCRIM: Quantifying State Criticality in Imitation Learning
The authors propose SCRIM, a state criticality measure that uses an adversarial perturbation policy and a value critic to estimate the worst-case degradation in downstream value caused by small perturbations to expert actions, scoring each demonstration state. These scores can be used to reweight behavior-cloning losses or to select compact, high-value subsets of demonstrations, reducing training data while maintaining performance. The method is evaluated on a variety of simulated and real robot manipulation tasks, and the paper was accepted to CoRL 2026.
arXiv cs.RO ·