More final project pitches + extra content

Thoughts

• I thought three of the papers we discussed were quite interesting:
• SDF collision detection for point-based cloth was very practical; it gave good results in real time and the math was pretty simple (just descend the distance field). I imagine it might not be suitable for very fine or dynamic geometry, though, since the rigid bodies the cloth collides with must be voxelized (in some manner) into the SDF. Either way, it enabled very sparsely sampled cloth faces to give respectable results, and played nicely with implicit surfaces defined by their SDF. I do wonder if (or when) the geometric representations we use in games/film/simulations will shift toward implicits rather than meshes.
• Muscle based locomotion: the hand-built neural feedback system for muscle based walking was very impressive - I didn’t think it was possible to get realistic results with such a simple hand-optimized system. Learned models end up running a very similar process in some sense, but with vastly more parameters and controls.
• I had seen the paper on genetically optimized muscle based locomotion before, but I again thought it was super cool. I didn’t previously know one of the simulation goals was matching a particular motion/velocity (and the initial configurations aren’t totally random), but that does make sense given the high dimensional parameter space. It was also cool that they could optimize muscle placements to sort of transfer-learn models onto different but similar body morphologies. Either way, I find physics-based locomotion more impressive than simply motion matching :^)