I am currently working on a classical fps game, I want to be able to generate larger numbers "enemies", ideally in the hundreds. Though this may not be feasible. I am hoping to use physics and Rigidbod[ies]
for these enemy models.
Obviously the nice solution would be to have mesh colliders
on all these models but I'm pretty sure that will affect performance. So my alternative idea was to instead set them up initially with box colliders while they are atleast X distance from the player and then once they enter that radius around the player to switch (via enabling and disabling) these box colliders to mesh colliders.
Is this possible to do and what are the implications? My initial tests caused the model to lose its collision detection with the floor, and so drop through the environment.
Are there are performance implications to switching (enabling/disabling) colliders/if this would even make a difference?
Lastly if none of the above, is there a better solution to this problem that is performant?
The most common practice I know is using multiple primitive colliders under the same rigidbody; I believe this is much more efficient than using mesh colliders, plus, as an added bonus you can even very easily handle special hits (headshots do x2 damage, leg shots slow enemies down) since you know exactly which collider was hit.
Here is an example of how they did it back in the day in Counter Strike
Notice that you will have to place each primitive collider correctly on the bone structure for it to move correctly during animations.