← work

CoRL 2026 · co-authored paper

Artificial Foveated Perception for Mitigating Shortcut Learning in Robotic Foundation Models

GitHub repository

Co-authored paper accepted to CoRL 2026. Artificial Foveated Perception guides robot policies toward task-relevant visual regions during fine-tuning, reducing reliance on background distractions. The approach improves generalization without changing the policy architecture or adding an AFP call during inference.