A Pseudocode for Unadversarial Example Generation Algorithm 1: Unadversarial patch generation Input: Pre-trained classifier with parameters w, loss function null

B.1 Overview of AirSim We conduct our simulation experiments using the high fidelity simulator, Microsoft AirSim. AirSim acts as a plugin to Unreal Engine, which is a AAA videogame engine providing access to high fidelity graphics features such as high resolution textures, realistic lighting, soft shadows etc. making it a good choice for rendering for computer vision applications. AirSim internally provides physics models for a quadrotor vehicle, which we leverage for performing autonomous drone landing. As a plugin, AirSim can be paired with any Unreal Engine environmnent to simulate autonomous vehicles that can be programmed with an API both in terms of planning/control as well as obtaining camera images. AirSim also allows for controlling environmental features such as time of day, dynamically adding/removing objects, changing object textures and so on.