Search

Autonomous flying robots, such as multirotors, often rely on deep learning models that makes predictions based on a camera image, e.g. for pose estimation. These models can predict surprising results if applied to input images outside the training domain. This fault can be exploited by adversarial attacks, for example, by computing small images, so-called adversarial patches, that can be placed in the environment to manipulate the neural network's prediction. We introduce flying adversarial patches, where multiple images are mounted on at least one other flying robot and therefore can be placed anywhere in the field of view of a victim multirotor. By introducing the attacker robots, the system is extended to an adversarial multi-robot system. For an effective attack, we compare three methods that simultaneously optimize multiple adversarial patches and their position in the input image. We show that our methods scale well with the number of adversarial patches. Moreover, we demonstrate physical flights with two robots, where we employ a novel attack policy that uses the computed adversarial patches to kidnap a robot that was supposed to follow a human.

Autonomous flying robots, e.g. multirotors, often rely on a neural network that makes predictions based on a camera image. These deep learning (DL) models can compute surprising results if applied to input images outside the training domain. Adversarial attacks exploit this fault, for example, by computing small images, so-called adversarial patches, that can be placed in the environment to manipulate the neural network's prediction. We introduce flying adversarial patches, where an image is mounted on another flying robot and therefore can be placed anywhere in the field of view of a victim multirotor. For an effective attack, we compare three methods that simultaneously optimize the adversarial patch and its position in the input image. We perform an empirical validation on a publicly available DL model and dataset for autonomous multirotors. Ultimately, our attacking multirotor would be able to gain full control over the motions of the victim multirotor.

Deep Neural Networks are widely applied for solving Computer Vision tasks for Unmanned Aerial Vehicles (UAVs). For some applications, the predictions of the neural networks (NNs) directly influence the motion planning or control of the UAVs. However, the neural networks are highly prone to adversarial attacks, which has a severe negative impact on the drone’s safe operation. With this work, we are planning to perform a physically realizable attack on a neural network analyzing camera images. The control of the UAV is directly influenced by the predictions of this NN. The generated adversarial attacks will be printed and attached as adversarial patches to an attacker UAV. By choosing which patch to present given the current relative poses of victim and attacker, the attacker will achieve full control over the victim UAV.

In recent years, many light-microscopy protocols have been published for visualization of nanoscale structures in the kidney. These protocols present researchers with new tools to evaluate both foot process anatomy and effacement, as well as protein distributions in foot processes, the slit diaphragm and in the glomerular basement membrane. However, these protocols either involve the application of different complicated super resolution microscopes or lengthy sample preparation protocols. Here, we present a fast and simple, five-hour long procedure for three-dimensional visualization of kidney morphology on all length scales. The protocol combines optical clearing and tissue expansion concepts to produce a mild swelling, sufficient for resolving nanoscale structures using a conventional confocal microscope. We show that the protocol can be applied to visualize a wide variety of pathologic features in both mouse and human kidneys. Thus, our fast and simple protocol can be beneficial for conventional microscopic evaluation of kidney tissue integrity both in research and possibly in future clinical routines., QC 20211215

The Paris climate agreement aims at holding global warming to well below 2 degrees Celsius and to “pursue efforts” to limit it to 1.5 degrees Celsius. To accomplish this, countries have submitted Intended Nationally Determined Contributions(INDCs) outlining their post-2020 climate action. Here we assess the effect of current INDCs on reducing aggregate greenhouse gas emissions, its implications for achieving the temperature objective of the Paris climate agreement, and potential options for overachievement. The INDCs collectively lower greenhouse gas emissions compared to where current policies stand, but still imply a median warming of 2.6–3.1 degrees Celsius by 2100. More can be achieved, because the agreement stipulates that targets for reducing greenhouse gas emissions are strengthened over time, both in ambition and scope. Substantial enhancement or over-delivery on current INDCs by additional national, sub-national and non-state actions is required to maintain a reasonable chance of meeting the target of keeping warming well below 2 degrees Celsius.