One of the big challenges for forward dynamics is getting your model to match motion capture recorded performances. There are two ways you can do this. The easy way is to start with a nicely matched model and calculate the forces that you need to move the model to the next recorded position. You then apply these forces and move the model along to the next frame. This is called local matching and because it works incrementally through the individual motion capture frames it is relatively quick to calculate. However the problem with this approach is that sometimes you need to have done something in an earlier frame to allow a movement to work in the current frame (this is the parallel parking problem). Similarly sometimes you need to allow some error in an earlier frame to get a much better match in later frames (there is always some error). To get around these difficulties we can use a different approach. This alternative is not to match each frame individually but instead to try and match the whole series of frames all at once. This is called global matching and is much more difficult because there are a lot more parameters to optimise. We have recently managed to get global matching working properly and you can see an example below.
