Extended Position Based Dynamics in PyTorch

Based on the StVk Garment model, we develop an cloth simulation toy example using pytorch.

Installation

This project is python and pytorch based and we've only tested it on ubuntu. Install dependecies with pip install -r requirements.txt

Usage

• simulate_global_stvk_bend implements a global position based dynamics update (using backward euler). This global update is typically slower to converge on single node updates.
• simulate_loss_stvk_bend treats the cloth simulation as a optimization problem where energies are defined as optimization target for the network to reach equilibirum.
• simulate_dynamic_stvk_bend will combine the previous two types of simulation techniques and try to introduce more dynamics into the process (preserves momentum through position based dynamics)
• visualize_open3d_animation provides a cross-platform visualization of the simulation result (tested on windows too).

python simulate_loss_stvk_bend.py # will output tshirt-drop#loss.npz
python simulate_global_stvk_bend.py # will output tshirt-drop#global.npz
python simulate_dynamic_stvk_bend.py # will output tshirt-drop#dynamic.npz

python visualize_open3d_animation.py "tshirt-drop#loss.npz" 10  # will render the stvk loss simulation in 10 fps (in terms of simulation) (6 times slower)
python visualize_open3d_animation.py "tshirt-drop#global.npz" 10  # will render the global xpbd simulation in 10 fps (in terms of simulation) (6 times slower)
python visualize_open3d_animation.py "tshirt-drop#dynamic.npz" 10  # will render the dynamic xpbd + loss simulation in 10 fps (in terms of simulation) (6 times slower)