Differentiable Physics Simulation for Learning and Robotics
, NVIDIA
Highly Rated
Learn about recent advances in differentiable physics simulation for GPUs. We'll cover the mathematical preliminaries of automatic differentiation, and show how to design a Python domain-specific language that can be used to generate high-performance CUDA code for computing simulation gradients. The presented framework can be applied to the simulation of rigid bodies, thin-shell, and solid finite element models. We'll show how to apply this framework to tasks such as parameter estimation and optimal control using the PyTorch framework.