I have a ragdoll setup and basically all I want is to instantiate a bunch of them, and have them fall onto a pile and collide with each other.

What I've done so far is I have a ragdoll solver -> joint deform -> copy to points, and feeding a single plane into the collision on the ragdoll solver I get this (video [imgur.com])

There's however no collisions between them, because the ragdoll solver is only taking in the single plane



I can't just feed the output of the copy to points back into the ragdoll solver as collision geometry, so it feels that I need a fundamentally different approach?

Using the crowd tools is probably the easiest way to simulate a large number of ragdolls. The KineFX ragdoll solver is aimed more at a single character, so it would be awkward to set up unless you merge all the skeletons together with unique joint names, or something like that