A novel point-based convolutional neural network architecture is proposed to effectively learn object dynamics from 3D point cloud or mesh data, capturing both within-object and between-object interactions through specialized convolution operators and a hierarchical U-Net structure.
Vertex Block Descent introduces a physics solver for elastic body dynamics with unconditional stability and exceptional performance.
Techniques for parallelizing large rigid body simulations to achieve scalability and load balancing.
Vertex block descent introduces a physics solver for elastic body dynamics, offering stability, performance, and convergence.