Zero-Shot Uncertainty-Aware Deployment of Simulation Trained Policies on Real-World Robots