Specifications
225
very quick and the system predicts very rapid changes of muscle activation, then the result may not be
realistic.
Another possible source of error is the distribution of force between the muscles. The body has more
muscles than strictly necessary to carry most loads, so is infinitely many different combinations of muscle
forces will balance the external loads. The way AnyBody picks the right one is by an optimality criterion. The
system presumes that the body wants to make the best of its resources. The user has some amount of
control over this criterion, but in its basic form it is a minimum fatigue criterion that distributes the loads as
evenly as possible between the muscles taking their individual strengths into account. Please refer to A
Study of Studies for more detailed information.
So the system basically presumes that the body has the knowledge and the desire to activate muscles
optimally. This is supported by a lot of research, but the precise criterion employed by the body is a matter
of continuous discussion. Furthermore, the ability to instantly choose the optimal muscle recruitment most
likely requires that the movement is skilled and that the required changes of muscle activation are not faster
than the electro-chemical process of muscle contraction can accommodate.
Errors in the software
All software has bugs, and very probably this is also the case for the AnyBody Modeling System. However, in
terms of muscle recruitment, the validity of the software was validated independently in 2004 in a Ph.D.
thesis by Erik Forster from the University of Ulm, Germany. The thesis is available from the list of
publications in the AnyBody Research Project, www.anybody.aau.dk/publications.htm
. The basic idea was to
program an independent special-purpose application for gait simulation and then compare it to an identical
gait model in AnyBody. If the results were identical, it would prove the correctness of the algorithms of both
systems. The result was that the output data of the two systems were identical on all but a tiny fraction of
the data. Closer investigation of this tiny fraction revealed that different algorithms - although
mathematically similar - can produce slightly deviating results due to round-off errors.
When compared to the modeling errors and approximations due to the recruitment assumptions, errors in
the software are much less likely to disturb the result of the computation significantly.
Methods of validation
The expression 'garbage in - garbage out' is very much valid for biomechanical simulation. The quality of the
output can never be better than the input. This means that the first step of any validation is to check the
quality of the input. Input comes in the form of movements and applied forces, where the former is the
more difficult. A rough check of the specified movements can be obtained by running a kinematic analysis
and charting the positions, velocities and above all accelerations of characteristic points and segments in the
model as illustrated for the rowing model above. Notice that proximal body parts tend to be heavier than
distal body parts, so larger accelerations are plausible in the distal parts.
Gravity = 9.81 m/s^2 is a good measure to compare your values to. If the accelerations oscillate or attain
unrealistic values, then the input positional information definitely needs careful reviewing and probably
smoothing with a low-pass filter.
Kinematic input
The picture below shows the thorax position variation over time recorded by digitizing images from a video
capture (model curtesy of the Technical University of Vienna) of ergometer rowing. The red curve is the
horizontal position, and the green curve is the vertical position. The lateral position remains zero and is the
blue curve. It looks very reasonable, and it does not seem to be infested with significant noise.