Differentiable Material Point Method’s Recent Connections to Real World

Differentiable physical simulations are increasingly essential tools in modern intelligent systems. Among various differentiable solvers, the Material Point Method (MPM) stands out due to its inherent differentiability. One of its key advantages over alternative differentiable simulators is its adeptness at handling topologically dynamic solid and fluid materials, as well as managing frictional contact. In this talk, we will explore recent contributions from my research group on leveraging differentiable MPM for multiple applications. These range from training controllers for soft robotic locomotion to optimizing the energy-efficient design of 3D-printed unmanned aerial vehicles, and reconstructing the geometry and physical properties of diverse materials from video footage.