Hello everyone !
I have an agent that is holding an object in his hand.
How would I randomize the wrist orientation slightly so they are not all exactly the same?

I found a way to do it, it seems to work on the crowdsource, but gets overwritten by the simulation.
From what I gather atm, the sim gets the clip that's driving the current state. Since the clip doesnt contain the randomized info, its all gone.
As it is, the randomization occurs after the crowdsource node so I've pipped the ‘TEST_RANDOM’ null into the sim, instead of crowdsource obj. It didnt break the sim, but no randomized wrists go thru either.

Any ideas please ?
Bedankt !
A

PS: Something that remains obscure to me is why we must define the clip twice, crowdsource and sim state. Im pretty ‘houdini fresh’ atm, but feels confusing, or ‘duplicated’ so to speak.

You probably want to do that post-sim (or in DOPs, after the crowd solver) - otherwise any changes you make to the agent's pose will be lost when the solver updates the agent's clip times.

For the second question, there isn't a 1-1 correspondence between clips and states. In the crowd source SOP you set what the initial state of the agent is, but a state can e.g. randomly select between multiple clips, and multiple states could use the same clip but with different POPs wired below for different behaviours. It's only in some cases where the state name matches the clip name.