I thought this was pretty neat and had some potentially interesting applications / ramifications - so I decided to share it. It's a muscle based locomotion / animation system that procedurally generates appropriate gaits for bipedal characters and allows them to compensate for external stimuli to a minor extent such as slopes or collision. I'd love to see this evolve further!
Video:
https://vimeo.com/79098420
White paper:
http://www.staff.science.uu.nl/~geijt101/papers/SA2013/SA2013.pdf
This showed up in my facebook news feed this morning. I'm especially impressed at how the little dinosaur developed a kangaroo-like leaping gait at high speeds.
The relevant paper is available here. I'm really tempted to try and implement something along these lines...
I wouldn't expect it to be useable at run time. I suspect the solver requires quite a few iterations to get it right, and that each iteration is in itself a processing intensive operation.
This is... wow! Just WOW!
Some years ago I was also thinking at stuff like that, if we eliminate the detail of how muscles generate forces, there could be a way to just solve the forces right? Well, it turns out it would have been quite some work!
The genericity of the algorithm is impressive.
I wouldn't expect it to be useable at run time.
The evolution of a viable controller is an offline process, almost certainly. Though given that their controllers seem to be robust in well under 1,000 generations, that's going to be a pretty quick offline process.
Once you have evolved a stable controller, applying it at runtime should be entirely feasible.
