Fig. 1From: Collision-aware interactive simulation using graph neural networksOverview of the proposed method. GNN is the base model. A collision-aware recursive regression module updates the network parameters recursively using interpenetration distances calculated from vertex-face and edge-edge tests. A novel self-supervised collision term (random latent space vector \({\overline{z}}_{rand}\sim\) (0, 1)) provides a more compact collision responseBack to article page