Abstract
Algorithms that fuse multiple input sources benefit from both complementaryand shared information. Shared information may provide robustness to faulty ornoisy inputs, which is indispensable for safety-critical applications likeself-driving cars. We investigate learning fusion algorithms that are robustagainst noise added to a single source. We first demonstrate that robustnessagainst single source noise is not guaranteed in a linear fusion model.Motivated by this discovery, two possible approaches are proposed to increaserobustness: a carefully designed loss with corresponding training algorithmsfor deep fusion models, and a simple convolutional fusion layer that has astructural advantage in dealing with noise. Experimental results show that bothtraining algorithms and our fusion layer make a deep fusion-based 3D objectdetector robust against noise applied to a single source, while preserving theoriginal performance on clean data.