We study the creation of physical adversarial examples, which are robust to real-world transformations, using a limited number of queries to the target black-box neural networks. We observe that robust models tend to be especially susceptible to foreground manipulations, which motivates our novel Foreground attack. We demonstrate that gradient priors are a useful signal for black-box attacks and therefore introduce...

This project in the field of robotics demonstrates the application of a neural network that uses the sinusoidal activation function for the task of perturbed walking. The usage of such networks is analysed. Furthermore, the most common local minima and the methods of resolving them are presented. The small number of neurons allows the network to be deployed on a...