Pi-0

The 'Pi-0' dataset/benchmark is used to evaluate the effectiveness of mechanistic interpretability methods for steering vision-language-action models by providing a structured environment for assessing model behavior and intervention strategies.

Papers using Pi-0 (2)

STRONG-VLA: Decoupled Robustness Learning for Vision-Language-Action Models under Multimodal Perturbations2026Mechanistic interpretability for steering vision-language-action models2025