What is the correct way to get the distance between and the corresponding closest points?

[BolunDai0216]

Hi,

I just started to look into this library, and I noticed that when computing the distance between two objects

import hppfcl
import numpy as np

box = hppfcl.Box(np.array([1.0, 1.0, 1.0]))
capsule = hppfcl.Capsule(1.0, 2.0)

M_box = hppfcl.Transform3f(np.eye(3), np.array([0, 0, 0]))
M_capsule = hppfcl.Transform3f(np.eye(3), np.array([10, 0, 0]))

req = hppfcl.DistanceRequest()
res = hppfcl.DistanceResult()

dis = hppfcl.distance(capsule, M_capsule, box, M_box, req, res)

It does not give any other information, like the corresponding closest points. Is possible to help me understand how I can obtain that information?

Thanks in advance!

[jcarpent]

Did you look at the res quantity? This is very similar to FCL here.

[BolunDai0216]

Thanks for the response!

So we can get the closest points using

capsule_point = res.getNearestPoint1()
box_point = res.getNearestPoint2()

This seems to give the points in world frame coordinates, is there a way to directly get the points in body frame coordinates?

Another question I have is, when I want to compute the distance between the same objects just with different Transform3fs, is there an update function for that?

[jcarpent]

This seems to give the points in world frame coordinates, is there a way to directly get the points in body frame coordinates?

Just apply the inverse transformation on the points.

Another question I have is, when I want to compute the distance between the same objects just with different Transform3fs, is there an update function for that?

Do a new call. You won't gain so much, except in specific scenarios.
What do you expect with the update?

And please do not forget to cite this repository in your code and publications.

[BolunDai0216]

Should I also look at the hppfcl.CollisionResult() instance for the distance and closest points? When I look at it

(Pdb) res_left
<hppfcl.hppfcl.CollisionResult object at 0x11e315810>
(Pdb) dir(res_left)
['__class__', '__delattr__', '__dict__', '__dir__', '__doc__', '__eq__', '__format__', '__ge__', '__getattribute__', '__gt__', '__hash__', '__init__', '__init_subclass__', '__le__', '__lt__', '__module__', '__ne__', '__new__', '__reduce__', '__reduce_ex__', '__repr__', '__setattr__', '__sizeof__', '__str__', '__subclasshook__', '__weakref__', 'addContact', 'cached_gjk_guess', 'cached_support_func_guess', 'clear', 'distance_lower_bound', 'getContact', 'getContacts', 'isCollision', 'numContacts', 'timings']

I did not see anything like distance or getNearestPoint.

[jcarpent]

distance_lower_bound is your distance info and getContact(0) will return the contact point if any as a hppfcl.Contactstruct.

[jcarpent]

@jcarpent I was expecting it to be faster, but it seems like the current version is fast enough.

Fast is a relative concept. What do you clearly mean here? Could you give real timings here?

[BolunDai0216]

What I meant was, if there exists an update(), then it would probably save time by not creating new variables at each time step. But using the distance function is fast enough for my purpose (less than 1 ms per call). I did not actually time it, but it was definitely faster than 1 ms.

[jcarpent]

Yes. You should be around the us scale