dmpbbo-glde

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:

• https://github.com/stulp/dmpbbo/blob/c73e1418b611f31cfbe652df51f14f8c18552897/dmpbbo/dmps/Dmp.py#L101

Contents of the repo

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:

Demos and illustrations

• 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 and presentation_optimization.py: Generate plots for the ICRA presentation.

Experiments

• 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.

Helper modules

• 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.