Steering Robustness into World Action Models via Mechanistic Interpretability and Optimal ControlPublished in Conference on Robot Learning (CoRL), 2026Share on Twitter Facebook LinkedIn Previous Next