Biblio
Learning-enabled components (LECs) are widely used in cyber-physical systems (CPS) since they can handle the uncertainty and variability of the environment and increase the level of autonomy. However, it has been shown that LECs such as deep neural networks (DNN) are not robust and adversarial examples can cause the model to make a false prediction. The paper considers the problem of efficiently detecting adversarial examples in LECs used for regression in CPS. The proposed approach is based on inductive conformal prediction and uses a regression model based on variational autoencoder. The architecture allows to take into consideration both the input and the neural network prediction for detecting adversarial, and more generally, out-of-distribution examples. We demonstrate the method using an advanced emergency braking system implemented in an open source simulator for self-driving cars where a DNN is used to estimate the distance to an obstacle. The simulation results show that the method can effectively detect adversarial examples with a short detection delay.
With self-driving cars making their way on to our roads, we ask not what it would take for them to gain acceptance among consumers, but what impact they may have on other drivers. How they will be perceived and whether they will be trusted will likely have a major effect on traffic flow and vehicular safety. This work first undertakes an exploratory factor analysis to validate a trust scale for human-robot interaction and shows how previously validated metrics and general trust theory support a more complete model of trust that has increased applicability in the driving domain. We experimentally test this expanded model in the context of human-automation interaction during simulated driving, revealing how using these dimensions uncovers significant biases within human-robot trust that may have particularly deleterious effects when it comes to sharing our future roads with automated vehicles.