>>Here we have simulated characters that learn to adapt motions to irregular terrain, where the original video clips were recorded from actors on flat ground.
That simulation looks quite good, though I'm struck that it has apparently not learned the simplest trick to success, launching your flip off of a higher point and landing lower
Do I detect a fellow partner acrobat ?
But that would be nice indeed.
The problem however is we work on contact area between each other (the 2 acrobat partners) that isn't perfectly stable. (For example handstand in hands.) Or there might be a way to abstract it away.
