LineCoSpar

This repository contains the source code for the LineCoSpar algorithm and implementation of the experiments described in the paper:

Human Preference-Based Learning for High-dimensional Optimization of Exoskeleton Walking Gaits.

Paper       Video

LineCoSpar is a human-in-the-loop preference-based algorithm that enables optimization over many parameters by iteratively exploring one-dimensional subspaces. LineCoSpar is an extension of CoSpar. The high-dimensional performance of LineCoSpar is demonstrated through both simulations and human-subject trials. The human-subject trials in the IROS publication consisted of optimizing 6 walking gait parameters of a lower-body exoskeleton for 6 able-bodied subjects. Our analysis of the lower-body exoskeleton experiments highlights differences in the utility functions underlying individual users' gait preferences.

---

Respository Contents

• exoskeleton_pref_learning.py: version of the LineCoSpar algorithm that was deployed on the lower-body exoskeleton for the IROS publication experiments.
• Plotting: code to plot the simulation results
• cartpole_experiments: code to run simulated experiments of LineCoSpar on a cart-pole system. This set of simulated experiments, which was not discussed in the IROS publication, provide another example of how the algorithm can be used.
• gaitAnalysis folder: code to fit various cost function terms to the lower-body exoskeleton experimental preferences. This folder also contains code to plot the CoM and CoP trajectories of the most- and least-preferred gaits.
• synthetic_fns: code for simulations on the synthetic functions used in the IROS publication.