I created a model based on the AAU Human of the AMMR of AnyBody. The Model performes a knee extension/flexion in a sitting position which gets angle and torque values from a dynamometer. I wanted to take a look at the muscle activity of the knee extensors (rectus femoris, vastus lateralis, vastus medialis). After the inverse dynamic analysis, the activity of rectus femoris seems logical, but vastus lateralis and medialis have a really low activity (e-05) which is impossible due to the given exercice. I tried to do a muscle calibration but it doesn’t seem to solve the problem.
The definition of the rectus femoris and vasti muscles (strength/attachments/path) come from the original TLEM dataset. As it happens right now the rectus femoris has a larger moment arm to perform the extension task, and is strong enough to be activated alone (think of the muscle recruitment). If the torque was larger the RF activation would peak and the muscle recruitment optimization would engage other muscles to help. So it might be that this definition is not perfectly accurate - some improvements could correct the situation.
So a couple of ways to solve it would be to check the definition (line of action) of the vasti muscles - which possibly (and quite likely) can be improved. An improvement in the calibration might make them stronger (or RF a little weaker) as well. Unfortunately I do not have an out of the box solution for you right now, but we will look into this issue for the next generation of the AMMRs.
Do you by chance use the polynomial criterion with the power of 2? Which recruits stronger muscles, whereas larger powers tend to use muscle synergies more (by default we use the power of 3 polynomial criterion).