I think ResNet pretty much solved vanishing gradients. As for exploding gradients, that is typically with good parameter initialization and normalization. The paper in question proposes an alternative to normalization.