B.Sc Thesis on Human Gait Synthesis using PredictiveSim

[Jakob Welner]

Hi,
I am sorry in advance if thise post is misplaced in this forum, but I have found no better options.

I recently finished my B.Sc Thesis in Mechanical Engineering revolving around Biologically Inspired Human Gait Synthesis using PredictiveSim by Tim Dorn and Jack Wang.

My thesis was a means to get myself acquainted with the topic of Predictive Simulation in biomechanics, starting from zero, and explaining all the building blocks one by one to the best of my capabilities.

I am posting it here in case someone can get anything useful out of it, as I struggled to find good learning material and would have appreciated something similar when I started.

Thesis PDF: https://www.dropbox.com/s/bpj7efrq8dr0nrn/main.pdf?dl=0

In case this inspires anyone to read my thesis who then finds any errors or have other comments I would be more than happy to hear about them or even better to start an open discussion about them. 

Cheers
- Jakob




A bit further details:

For those who don't know PredictiveSim it is a freely available software implementing Predictive Simulation for Biologically Inspired Human Gait, made, as far as I know, as part of the publication: "Predictive Simulation Generates Human Adaptations during Loaded and Inclined Walking" by Tim Dorn, Jack Wang, Jennifer Hicks and Scott Delp.

The thesis is generally explorative and starting from the very basics as I had no former knowledge on the subject. I furthermore went wide instead of deep as I wanted to touch upon as many techniques and theories as I could fit in, in order to form a better basis for my eventual masters degree. The fact that I had no former knowledge on any of this may show in some rather rudimentary observations and points being made, but I guess that could serve as a helping hand for others who are just starting out as well.

Thesis content:
- Overview of Predictive Simulation, describing submodels and their interplay (musculoskeletal system, optimizer, controller, simulator, objective function)

- Overview of the locomotor nervous system including inner priorities (Objective function), muscle reflexes and supraspinal control

- PredictiveSim implementation of the above mentioned submodels

- Custom Implementation of additional features to the software including: 
- - Blind Random Search as an alternative optimization algorithm, 
- - Perturbations
- - Simple 'supraspinal control' implemented simply by allowing to swap paramter datasets during simulation. 

- Experiments including optimization for varying speeds and perturbations with subsequent PCA for correlating control parameters to their resulting speed/stability. This was done both for independent optimizations and sequential where both were optimized with varying target speed and perturbation but where the former used the same initial condition and the latter were sorted and each new optimization took the previous result as its initial condition. This was done because each and every optimization appeared to land on a new local minima, making it difficult to compare parameters between different 'strategies'. Performing sequential optimizations made it possible to stay more or less withing the same 'movement strategy' while slightly varying speed and/or perturbation, thus allowing easier analysis. Also, it turned out that when exploring supraspinal control there was a difference between changing speed with parameter sets from the same 'strategy' rather than from separate 'strategies'.
A larger acceleration could be performed without stumbling when changing between control parameters for varying speed when they were of the same 'strategy' rather than if the parameters came from an independent optimization, despite having the same objetive parameters.

[Dimitar Stanev]

Hi,

I would encourage you to create a SimTK project and upload any file that you feel uploading related to your thesis so that others can read and reproduce what you have done.

Best

[Jakob Welner]

Ah yes, of course. I will do that. Thank you for the heads up.

Cheers
-jakob