sergiorivera50 / bayesian-crutch-geometries

This repository presents our research on optimizing crutch designs using Gaussian Processes (GPs) and Bayesian Optimization (BO). We introduce a novel loss function that blends subjective (pain, instability, effort) and objective measures, leading to a personalized, more efficient, and comfortable crutch design.

Personalising Crutch Geometries through Bayesian Optimisation

Authors

Riccardo Conci ([email protected])

Riccardo Ali ([email protected])

Deepro Choudhury ([email protected])

Sergio Rivera ([email protected])

Abstract

Crutches are optimised for stable motion, but this safety comes at the cost of comfort and speed. In this paper, we employ Gaussian Processes (GPs) and Bayesian Optimisation (BO) as hypothesis generators to find better crutch configurations, which we validate on a physical prototype. We do so by defining a novel loss function indicating the quality of a crutch design which combines subjective metrics (joint pain, instability and effort) and the corresponding objective ones.

Finally, we (1) use this methodology to build a more stable, less effortful and less painful personalised crutch design and (2) use the knowledge built by the GP through these experiments to enhance our understanding of the physical dynamics of crutching.

How to use this repo

Most of the work is contained within main.ipynb, where you will find our setup and all the experiments.

Feel free to read the full report at report.pdf.