Inverse distance weighting to rapidly generate large simulation datasets.

Area of Science:

• Biomechanics
• Machine Learning
• Computational Modeling

Background:

• Generating large biomechanical datasets for machine learning is challenging.
• Physics-based simulations like computed muscle control (CMC) are computationally intensive.
• Interpolation methods offer a faster alternative for data generation.

Purpose of the Study:

• To evaluate inverse distance weighting (IDW) as a low-cost and accurate method for interpolating muscle activations from CMC simulations.
• To assess the accuracy and computational efficiency of IDW compared to CMC.

Main Methods:

• IDW was applied to muscle activation data from OpenSim lateral pinch simulations.
• Simulated data were organized into grids of varying sparsity.
• Muscle activations were calculated using both CMC and IDW for random mass-height pairs.
• Interpolation errors and their impact on pinch force were evaluated.

Main Results:

• Root mean square error for interpolated muscle activations decreased with increasing grid density, not exceeding 4%.
• IDW interpolation errors minimally impacted simulated lateral pinch force (≤0.1 N).
• IDW computation time (0.95 core-seconds) was significantly faster than CMC (4.22 core-minutes).

Conclusions:

• IDW is a practical and efficient method for estimating muscle activations from sparse CMC datasets.
• This approach can accelerate the generation of biomechanical data for machine learning applications.
• Future research can extend IDW to diverse biomechanical tasks and populations.