Add support for Spot with arm #57
Conversation
|
@zkingston this does not appear to be a Mac-only issue; I can reproduce it on Ubuntu. Additionally, I'm seeing some sort of hang in More specifically: only for |
|
I fixed the RNG issue (undefined behavior during access beyond bounds of a SIMD vector when converting to a float array), and the Spot can now plan! I also found that the observed hangs seem to come from an issue with RRTC going to a huge number of iterations - my running theory is that we're hitting the limits of the nearest neighbors structure, given the number of samples + the high dimensionality of the Spot relative to other supported robots, but I'm not sure about this. The need to take so many iterations (300k+) is likely due to the naive guesses I used for the Spot's validation fineness and RRTC range, or due to the SRDF used to generate the Spot's fkcc missing some important disallowed collisions. Still, I think this support is OK for now, maybe modulo @zkingston tuning of the RRTC range etc. I'll see if I can get a more refined SRDF. |
|
As an aside, I'm not sure why the format check is failing? I've run |
|
I've added support for a variant of the Spot with only its arm left free - this seems to work reliably (based on a bit of |
claytonwramsey
left a comment
claytonwramsey
left a comment
There was a problem hiding this comment.
all looks good to me!
you weren't lying about compilation times though:
Successfully installed vamp-planner-0.6
________________________________________________________
Executed in 141.42 secs fish external
usr time 320.50 secs 0.00 micros 320.50 secs
sys time 10.50 secs 396.00 micros 10.50 secs
There was a problem hiding this comment.
may want to pick out these changes as a separate PR since that will be way easier to merge
There was a problem hiding this comment.
should we remove panda_sphere_cage since this is a very similar experiment? Or leave it since it's the easiest example to reproduce?
This PR adds support for the Boston Dynamics Spot robot, based on the URDF files available here: https://github.com/bdaiinstitute/spot_description.
It also includes some minor fixes to the
random_dance.pyscript for compatibility with the latest VAMP.There's one issue to resolve before merging: for some reason, at least on ARM (MacBook Air M2), Halton sampling appears to be buggy (not only for Spot) and only returns non-zero values for the first 4 degrees of freedom.
We should resolve this issue just in case it's somehow caused by this PR (though I doubt it - I more suspect a Mac issue or a regression due to the recent change in robot kinematics generation backend).