We introduce a novel object selection mechanism to allow sim-trained policies to rapidly adapt to real-world visual perturbations.