I can't seem to find any documentation regarding the measurement of 'composite motion' and how it might be calculated.

The graphs seem to indicate that it is in mm, so I would guess that it's something similar to Framewise Displacement (Power et al., 2012).  Is this the case? 

Best,
Jonathan O'Rawe

Lines 432-441 in the code describe the movement parameter computations.  My understanding is that the composite motion you mention is the Euclidean norm for the regressors - hence it being a "summary" statistic.  I think the specific calculation is on line 458. (I may have added lines while looking through the code, so these are close, but may be off by a couple.)
HTH,
Brianne

Hi. I am stealing something from the Conn forum that might address your question (sorry it's a year later!)
Best,
Mary 

Nov 9, 2015 07:11 PM
Alfonso Nieto-Castanon
McGovern Institute for Brain Research. MIT RE: ART output
Hi David,
It is similar to framewise displacement but not exactly the same ("framewise displacement" converts angular differences to mm by multiplying by a constant factor -projecting to a sphere-, and then sums the 6 individual translation/rotation displacement absolute measures; ART's "composite motion" measure estimates instead the maximum voxel displacement resulting from the combined effect of the individual translation and rotation displacement measures)
Hope this helps
Alfonso