S^2 Metric for Evaluating Embodied Policy via Sensorimotor Alignment
October 22, 2026
S^2 is a training-free offline metric that measures the alignment between a model's sensory representations and the geometry of physical motion. The metric correlates positively with downstream robot task success, providing a low-cost proxy for closed-loop evaluation.
HOW THIS AFFECTS YOU
●
researcherYou can use this to evaluate robot foundation models without expensive physical rollouts.