Daniel Freeman
Daniel Freeman
Could you plot the gradient norm as a function of trajectory length for, say, a random policy? We've seen this kind of thing before, and it usually reduces to chaotic/unstable...
Hmmm, could you post some gifs of what happens during the collision? (alternatively, if you have a colab repro, that would be best :D )
Great, thank you! Yeah this is definitely not intended behavior. We'll get back to you shortly.
(Following up that we're still looking into this, we think there may be an offset not being propagated properly.)
Because we're using "impulsive" torques, these formulas actually end up being the same. But you're totally right that this is the more general expression for the rate of change of...
Yep, you're totally right o-Oscar. Managed to confuse the heck out of myself, because I know I've measured conservation laws for our codebase, and everything seemed to check out. Turns...
We have a fix for this internally, will ship out externally shortly. In short, `NAME_FIELDS` in component_editory.py should be this: ``` NAME_FIELDS = { 'bodies': ('name',), 'joints': ( 'name', 'parent',...
Hey Benjamin! I don't suppose you happen to have a repro for this, do you?
We haven't actually encountered environments that _needed_ more than the initial randomness you can cache into, say, 2048 environments, but your situation seems like a good test case. (see discussion...
I probably wouldn't reuse rng1. It really isn't that expensive to just generate another rng seed: ``` rng, rng1, rng2, rng3 = jp.random_split(rng, 4) ``` and then use rng3 for...