This repository contains the data and configuration files for our work "Soft Pneumatic Actuator Design using Differentiable Simulation". For supplementary material, please see here. The main implementation for the paper is contributed to PolyFEM, while the cascaded optimization is handled by the python script cascaded_optimization.py.
First, build PolyFEM and download (or build) MMG, noting down the directories where the binaries are located. Next, navigate to the directory where you want the optimization files to be saved to. You might want to do this in a tmux window as the optimization can take a long time. Then, run the following command:
python /path/to/pneumatic-actuator-design/cascaded_optimization.py --opt_example EXAMPLE_NAME --polyfem_build_dir /path/to/polyfem/ --mmg_build_dir /path/to/mmg/
where EXAMPLE_NAME is one of
- frog_quasistatic
- finger
- gripper_bellows
- worm
For some of the ablation studies, you can run
- gripper_bellows_shape_inside (gripper optimization using only target matching, no contact force maximization)
- frog_base (optimizatation of frog in vertex space)
- frog_quasistatic_base_weights_adjust (same as above, but starting with 4x the smoothing weights and decreasing over the optimization)
The 2D walker example is a simple optimization, so it can just be run by
/path/to/PolyFEM_bin --json /path/to/pneumatic-actuator-design/walker/run_walker.json --max_threads 8
The PolyFEM library automatically downloads its dependencies with cmake. The default linear solver in simulations is Eigen::PardisoLDLT, which requires MKL. This is not available on MacOS with Apple silicon, in which case Eigen::AccelerateLDLT or Eigen::CholmodSimplicialLDLT are recommended.
The python script requires meshio for doing IO with meshes and the libigl python bindings.
The log file in each output directory details the convergence of the forward problem (Newton iterations, convergence, timing on all steps) and the inverse problem, including energy and gradient norm.
The simulation results are exported as VTK format and can be visualized in Paraview. The optimization result files have the format of opt_state_{i}_iter_{j}, where i is the ID of the simulation (in the case where multiple simulations are included), j is the iteration number of the optimization. Once a cascade level is finished, the files are written as opt_{i}_{j}_{k} where i is the cascade step, j is the iteration number within the cascade step and k is the number of control points for that cascade level. You can rename all of the output files with the same name and increasing iteration number with the following command in the output directory
python /path/to/pneumatic-actuator-design/copy_output_sequential.py --opt_example EXAMPLE_NAME
This should save the files as opt_sequential_{i} where i is the iteration number. These can then all be loaded in the same Paraview import operation and they appear as different time steps, so you can examine the optimization by scrubbing the visualization.
If you'd like to setup a new optimization, please see our tutorial here (draft in progress).
Due to copyright limitations, we cannot distribute the models for the following files:
frog/frog_base_mesh.msh
finger/finger_base_mesh.msh
finger/finger_target_larger.obj
The base models for these files can be found at:
frog: https://www.turbosquid.com/FullPreview/1508446
finger: https://www.turbosquid.com/FullPreview/1103552
If you acquire these licenses, we can release the aforementioned models for the optimization via email: [email protected].
If you use this work/data, please cite our paper:
@inproceedings{10.1145/3641519.3657467,
author = {Gjoka, Arvi and Knoop, Espen and B\"{a}cher, Moritz and Zorin, Denis and Panozzo, Daniele},
title = {Soft Pneumatic Actuator Design using Differentiable Simulation},
year = {2024},
isbn = {9798400705250},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3641519.3657467},
doi = {10.1145/3641519.3657467},
abstract = {We propose a computational design pipeline for pneumatically-actuated soft robots interacting with their environment through contact. We optimize the shape of the robot with a shape optimization approach, using a physically-accurate high-order finite element model for the forward simulation. Our approach enables fine-grained control over both deformation and contact forces by optimizing the shape of internal cavities, which we exploit to design pneumatically-actuated robots that can assume user-prescribed poses, or apply user-controlled forces. We demonstrate the efficacy of our method on two artistic and two functional examples.},
booktitle = {Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers '24},
articleno = {106},
numpages = {11},
keywords = {Differentiable Simulation, Finite Element Method, Pneumatic Actuator, Shape Optimization, Soft Robotics},
location = {Denver, CO, USA},
series = {SIGGRAPH Conference Papers '24}
}