Increasingly, musculoskeletal models are being used with optimization tools to answer a variety of research questions. When these optimization problems require forward simulation of the models it is helpful if the simulations execute as fast as possible to reduce the time to obtain results, because optimization methods can require millions of distinct simulations and forward simulations of musculoskeletal models may even still require real time simulation times.

Many of the optimization methods used today wrap simulations based on generic physics engines which solve for the equations of motion at each time step. These physics engines are optimized for speedy forward simulation execution in a variety of ways, but because of their generic-ness they are unlikely optimized for the particular model. Many early prominent mutlibody dynamics codes, and some modern ones, use computer algebra systems (CAS) to symbolically derive long form analytic expressions for the equations of motion of the system. This allows for very high level mathematical descriptors of the equations of motion that can embody assumptions about the mathematical variables and expressions. With this high level description of the math it is possible to generate model and computer architecture specific low level codes through tailored code generation and compilation. This process has the potential to give significant speedups relative to the generic physics engines.

The hypothesis is that code generated from the symbolic forms can be significantly faster than a numeric based system. There has been work over the years in the SymPy and PyDy code generation tools that can be applied to musculoskeletal models.

Possible goals related to this topic (there is enough for multiple MSc's depending on the path):

• Improve SymPy's code generation facilities for arbitrary ordinary differential equations and differential algebraic equations that considers high level mathematical assumptions for optimizing the generated code.
• Create one or more benchmark models based on complex and typically slow-to-simulate biomechanics models using different software and compare execution times.
• Compare optimized forward simulations of different musculoskeletal simulation tools to tailored code generation from symbolic forms.
• Improve the Autolev to SymPy parser and get it working smoothly with complex Autolev models: https://docs.sympy.org/latest/modules/physics/mechanics/autolev_parser.html
• Implement the Featherstone, Jain, or similar equation of motion method in sympy.physics.mechanics, symbolically or as numeric code generation pattern.
• Implement code generation for typical models that slow simulations: stiffness, friction, switching, non-linearities, contact, muscle models, etc.
• Explore applicable parallelization techniques and apply to a complex muscle skeletal simulation.

How To Apply

If you would like to apply for this project, please send an approximately half page letter explaining your motivations and interest in the lab and project, CV or resume, a list of courses you've taken, the name of your MSc track, and any other relevant information to j.k.moore@tudelft.nl.