This PR fixes several bugs related to the ANSUR data and the anthropometric prior:
1) We were conditioning on columns ("Weightlbs" and "Heightin") that were SELF REPORTED and had the systematic bias you'd expect compared to the real thing. This is now fixed in test_MarkerFitter. These changes to the conditioning columns when setting up the anthropometric prior need to be replicated in engine.py
2) The measurement points for the anthropometric data were slightly off on the ANSUR_metrics.xml file (they'd never been systematically checked before). This has been carefully adjusted, and also needs to be replicated over in AddBiomechanics
3) As a result of all this, I was able to dramatically reduce the weight on the height prior, from 1e3 to 0.1, and still get good height convergence.
Also, this seems to help with pelvis roll quite a bit on the synthetic subject18 data!
nickbianco
commented
1 year ago
This PR fixes several bugs related to the ANSUR data and the anthropometric prior:
1) We were conditioning on columns ("Weightlbs" and "Heightin") that were SELF REPORTED and had the systematic bias you'd expect compared to the real thing. This is now fixed in
test_MarkerFitter. These changes to the conditioning columns when setting up the anthropometric prior need to be replicated inengine.py2) The measurement points for the anthropometric data were slightly off on theANSUR_metrics.xmlfile (they'd never been systematically checked before). This has been carefully adjusted, and also needs to be replicated over in AddBiomechanics 3) As a result of all this, I was able to dramatically reduce the weight on the height prior, from 1e3 to 0.1, and still get good height convergence.Also, this seems to help with pelvis roll quite a bit on the synthetic subject18 data!