Estimating the uncertainty of predictions is a crucial ability for robots in unstructured environments. Most mobile robots for indoor use rely on 2D laser scanners for localization, mapping and navigation. These sensors, however, cannot detect transparent surfaces or measure the full occupancy of complex objects such as tables. Deep Neural Networks have recently been proposed to overcome this limitation by learning to estimate object occupancy. These estimates are nevertheless subject to noise, making the evaluation of their confidence an important issue. In this work we study uncertainty estimation in deep models, proposing a novel solution based on a fully convolutional autoencoder. The proposed architecture is not restricted by the assumption that the uncertainty follows a Gaussian model, as in the case of many popular solutions for deep model uncertainty estimation, e.g., MC Dropout. We present results showing that uncertainty over obstacle distances is actually better modeled with a Laplace distribution. As an example of application where uncertainty evaluation is crucial, we also present an algorithm to build a map that includes information over obstacle distance estimates while taking into account the level of uncertainty in each estimate. We finally show how the constructed map can be used to increase global navigation safety by planning trajectories which avoid areas of high uncertainty, enabling higher autonomy for mobile robots in indoor settings.
View on arXiv