Repo for the experiments ICRA 2024 paper "Fitting Parameters of Linear Dynamical Systems\to Regularize Forcing Terms in Dynamical Movement Primitives" in which a novel DMP formulation is proposed.
Running this code requires v2.1 of dmpbbo to be installed: https://github.com/stulp/dmpbbo/releases/tag/v2.1.0
The main functionality for the novel formulation became available in V2.1.0 of dmpbbo, see for instance:
This repo provides experiments which apply this novel formulation to different datasets, which can be found in the data/ directory.
The following Python scripts are provided:
illustrate_from_ijspeert_to_kulvicius.py
: Show the effects of different DMP formulations on the distribution of function approximator parameters.illustrate_richards.py
: Illustrate the generalized logisitics function, also known as Richard's function.illustrate_novel_dyn_systems.py
: Illustrate the novel formulation with the generalized logisitics function.demo_optimize_dyn_sys_parameters.py
: A generic script for optimizing the parameters of the dynamical systems in a DMP.presentation_training.py
andpresentation_optimization.py
: Generate plots for the ICRA presentation.
experiment_bbo_of_dyn_systems.py
: Experiment 1 from the paper, i.e. train the different formulations on different datasets.experiment_optimize_contextual_dmp.py
: Experiment 2 from the paper, i.e. train a contextual DMP with the different underlying formulations on the coathanger dataset.experiment_from_ijspeert_to_kulvicius.py
: Bonus experiment: visualize more extensively the impact of the different formulations.
load_data_coathanger.py
: Load the coathanger data.save_plot.py
: Module with convenience function for saving a plot to SVG or other formats.colorpallette.py
: Some default colors.utils.py
: Various utility functions.