Update
We have been working and studying a lot, but without a lot of success so far.

We have been exploring machine learning for animation, based on the many scientific papers on the subject like: https://xbpeng.github.io/projects/DeepMimic/index.html and others. We have spent a lot of time trying to repeat some of the results to make a physical rag doll imitate an animation sequence while keeping balance, but the results so far have been discouraging. The video shows the beginning of training, then after a few hours and after a day. It is very time consuming to tweak parameters and waiting days to see the results. So far we haven't achieved much more than some drunken zombies. This technology will probably be available for everybody in the future, but for now we're on our own. We will keep on trying for a while, but we may have to give up at some point and find a different approach.

We have also been working on a system to automatically scale clothes to fit any girl at runtime, so we don't need to fit every dress and store it for each girl. We have hoped to be able to make this into a release with a variety of new clothes and it has been working to some extent, but not good enough to make a release. The video shows some of the problems, but it is more obvious in VR. We will have to get back to this some other time.

While waiting for the machine leaning to progress, we have started to work on some soft body physics.