This GitHub repository is presented as supplementary material to the paper Active Learning Design: Modeling Force Output for Axisymmetric Soft Pneumatic Actuators, authored by Gregory M. Campbell, Gentian Muhaxheri, Leonardo Ferreira Guilhoto, Christian D. Santangelo, Paris Perdikaris, James Pikul, and Mark Yim.

The pre-print is currently available on arXiv at: https://arxiv.org/abs/2504.01156. If you would like to cite this work, please use the reference below (in BibTeX):

@misc{campbell2025activelearningdesignmodeling,
  title={Active Learning Design: Modeling Force Output for Axisymmetric Soft Pneumatic Actuators}, 
  author={Gregory M. Campbell and Gentian Muhaxheri and Leonardo Ferreira Guilhoto and Christian D. Santangelo and Paris Perdikaris and James Pikul and Mark Yim},
  year={2025},
  eprint={2504.01156},
  archivePrefix={arXiv},
  primaryClass={cs.RO},
  url={https://arxiv.org/abs/2504.01156}, 
}

• Root files include .pkl files of characterization (model_data_dictionary) and test dataset (mass_lift_data_dictionary). It also includes general calculations and figures used in the paper.

• Data Acquisition includes files relevant to performing characterization and mass testing, including serial communication with relevant firmware.

• Hyperparameter Ablations includes results of Weights and Biases ablation study.

• Membrane Simulation includes theory-based simulations of membrane expansion and analysis of the results.

• model_data_dictionary_N_Pa.pkl: 'pickle' file containing a dictionary of membrane characterization data for 70mm radius Ecoflex 00-30 membranes. Each 'key' contains design parameter data in the form: thickness [mm], contact radius [mm], ring data [mm] (radius_1, width_1, radius_2, width_2). Each 'value' contains a tuple of arrays of membrane test data in the form ((us, ys), Fs, ws) where 'us' [n,7] represent contact height [mm], thickness [mm], contact radius [mm], ring data[mm], 'ys' [n,1] represent pressure [Pa], 'Fs' [n,1] represent contact force [N], and 'ws' represent weights to evenly weight each trial as opposed to each datapoint.

• mass_lift_data_dictionary_N_Pa.pkl: 'pickle' file containing a dictionary of membrane test data for 70mm radius Ecoflex 00-30 membranes. Each 'key' contains design parameter data in the form: thickness [mm], contact radius [mm], ring data [mm] (radius_1, width_1, radius_2, width_2). Each 'value' contains a tuple of arrays of membrane test data in the form ((us, ys), Fs) where 'us' [n,7] represent contact height [mm], thickness [mm], contact radius [mm], ring data[mm], 'ys' [n,1] represent pressure [Pa], 'Fs' [n,1] represent force from mass being lifted [N].

• acquisition.py - acquisition functions for NN model

• archs.py - NN architectures using Flax library.

• design_optimization.py - helper functions & optimization for NN model.

• figure_plotting.ipynb - solving for membrane test results reported in paper - creation of paper-relevant and video-relevant figures.

• membrane_optimization.ipynb - solving for height-maximization membranes.

• model-baseline_compare - using scikit-learn for k-fold cross-validation to baseline our NN model results. Solving ringless-only NN model error with k-fold cross-validation.

• select_next_experiment.ipynb - using model uncertainty to inform choice of next experiments (active learning).

• train-utils.py - NN model class and data-loader.

• Data_Manage.py - helper functions for interacting with .csv files.

• Run_Experiment.py - executable for running an automated set of data-characterization.

• Run_Mass_Experiment.py - executable for running a set of membrane mass testing.

• Serial_Comms.py - helper functions for interfacing with microcontroller via serial cummunication.

• Test_Procedure.py - helpfer functions for experiments.

• ablation_results.ipynb - explanation of hyperparameters with viwewing of dataframe and best parameters.

• wandb_export_monotonic_actuator_hyperparameter_ablations.csv - data from hyperparameter sweep from Weights and Biases.

Membrane Simulation:

• membrane_simulation.py - functions for executing theoretical simulation.

• sim_exp_compare.py - viewing simulation results and error relative to characterization data.

• sim_membranes.pkl - 'pickle' file containing a list of simulated test results. Each list item represents a simulation result and is of the form [name, [pressure [pa], force [N], height [m]] where 'name' [6,1] is of the form [thickness [mm], contact radius [mm], ring data [mm] (radius_1, width_1, radius_2, width_2)].

• sim_results.ipynb - completing simulations for the chaaracterized membranes. Note not all 22 membranes were successfully simulated. Note this script is likely to throw errors, as even the authors only found success of certain PC's. The data from these simulations is found in sim_membranes.pkl.