Hey ! I’m new to Unreal and I feel like there might be several way to solve my problem so I seek for guidance here.
I would like to use machine learning to animate a pawn which is physically simulated. Note that my pawn is way simpler than a human to reduce complexity.
I try to use Unreal systems as much as possible to avoid, you know, reinventing the wheel. So far I have a skeletal mesh with a physic asset that behave like a ragdoll on simulation. Now I need to simulate the strength that “muscle” would apply on different body part to allow my algorithm to control my pawn.
My best guess so far is to use angular motors on constraint of the physic asset and try to set some angular position as target from code. I believe it should produce movement.
Anyway it’s just a guess so if you guys have any idea or suggestion about how to solve this issue, or any other tip that could help, I would be glad to hear about it.
It’s basically an Either OR situation.
IF you move a joint/change bone rotations you won’t really have physics.
IF you have physics, you can’t really alter a joint.
Best exception to this you can try is the physical animation component - however, note that animations have a magnitude of power that far exceeds most objects, so an arm swing or a walk will cause most objects colliding with it to go flying as if an atomic bomb had just impacted near them… Altering rotations or whatever at the bone level, like an animation, will almost always out-do any force as well…
making your own thing, sure. nothing would stop you from checking other object/collisions before moving a bone in whatever way you see fit.
However This engine is not a good idea, with chaos being so poorly performant and PhysX being at least a version behind the last release before they killed it.
Doing your own physics calculation - would be a good exercise but like pointless - depends on final performance vs what you need it for after-all…