In this paper, we present an Assertion-based Multi-View Fusion network\n(AMVNet) for LiDAR semantic segmentation which aggregates the semantic features\nof individual projection-based networks using late fusion. Given class scores\nfrom different projection-based networks, we perform assertion-guided point\nsampling on score disagreements and pass a set of point-level features for each\nsampled point to a simple point head which refines the predictions. This\nmodular-and-hierarchical late fusion approach provides the flexibility of\nhaving two independent networks with a minor overhead from a light-weight\nnetwork. Such approaches are desirable for robotic systems, e.g. autonomous\nvehicles, for which the computational and memory resources are often limited.\nExtensive experiments show that AMVNet achieves state-of-the-art results in\nboth the SemanticKITTI and nuScenes benchmark datasets and that our approach\noutperforms the baseline method of combining the class scores of the\nprojection-based networks.\n