Huh, this is really interesting and in my field of research. Most people are using OpenSIM or closed source alternatives like Visual3d to do inverse kinematics/kinetics and there isn't much talk about machine learning in these fields, well, for this purpose, anyway.
To see the deadlift animation, and to have had it trained in a reinforcement learning is really interesting. Seems like this kind of simulation might drive the next wave of figure and trick development in many sports.
From figure skating to freestyle snowboard, many athletes are pushing the reslm of what’s possible by trial and error. What if you could get a bunch of simulations, and learn from its form.
I cannot wait to see this applied to BJJ, maybe the most complex sport in the world in terms of body mechanics. What amazing new techniqies could be uncovered if we had proper models.
Funny to see this as one of the first replies, it's something I've been thinking about lately as well. As long as there is basic collision detection in the model it would be hilarious to see what happens now if you put a couple of them in a proverbial cage. Probably 5-10 years before there's enough fidelity to actually innovate, but there might be enough coming from it to inspire new ideas before then.
While graphics often borders on photorealistic, animation still looks bad. Decades pass and even top of the line mocap still feels way off for me in the context of game environments.
Looks cool, but there doesn't seem to be a corresponding scientific publication that explains the inner workings and which shows that this result is generally useful outside of gaming/entertainment.
Very excited to see the direction this is going.