More discussion of mocap animation and research, started covering IK.

Thoughts

• It’s too bad we won’t be meeting to visit the panoptic studio on campus this semester; it was quite impressive.
• The research on pose capture from simpler input data (e.g. monocular rgb video) was also interesting, although all the learning based methods seem pretty similar to me.
• We also talked a bit more about learned motion capture sequencing, which I again thought was very impressive. I wonder if these technologies will soon show up in games, since the limitations of this work aren’t so bad in the context of games, and it has the potential to be far more immersive than hand-authored animation behavior trees. Plus, modern GPUs have hardware for neural inference.
• I hadn’t heard of the cyclic coordinate descent IK algorithm before (only inverse jacobian), so it was cool to see that such a simple procedure could converge to good results. I also wonder if it’s feasible nowadays to analytically solve (relatively small) IK problems, since I know that has been used for 2/3 joint cases in the past.
• The spline-based motion editing technique and cartoon convolution filter were also surprisingly simple but apparently quite effective - they both seem like something I could easily implement in Scotty3D.