extending capsule-capsule distance computation

[florent-lamiraux]

I have re-edited pull request #18 in order to highlight real differences against re-indentations.
Some tests fail on my computer (ubuntu 12.04 32bits), but they also fail with the official version.

[Karsten1987]

Thanks for your work!
Apologizes for my pull request, it was a mess.

However, the latest checkout of your repository doesn't contain the test_file:
https://github.com/Karsten1987/fcl/blob/master/test/test_fcl_capsule_capsule.cpp

With this it builds and this test runs without errors.

[florent-lamiraux]

Sorry for the missing file. This is fixed

[isucan]

Thanks guys!

[panjia1983]

I have fixed some bugs, and now all the tests passed on my machine before this commit. Maybe there are some bugs in this commit. (please apply my commit 293 and 294).

[florent-lamiraux]

I am running ubuntu 12-04 32bits with gcc (Ubuntu/Linaro 4.6.3-1ubuntu5) 4.6.3 and here are the results I get with the latest commits (6047de0):

florent@florent-laptop:~/devel/hpp/src/fcl/build$ make test
Running tests...
Test project /home/florent/devel/hpp/src/fcl/build
      Start  1: test_fcl_collision
 1/13 Test  #1: test_fcl_collision ................   Passed   33.20 sec
      Start  2: test_fcl_distance
 2/13 Test  #2: test_fcl_distance .................   Passed   26.98 sec
      Start  3: test_fcl_geometric_shapes
 3/13 Test  #3: test_fcl_geometric_shapes .........   Passed    0.08 sec
      Start  4: test_fcl_broadphase
 4/13 Test  #4: test_fcl_broadphase ...............   Passed  680.18 sec
      Start  5: test_fcl_shape_mesh_consistency
 5/13 Test  #5: test_fcl_shape_mesh_consistency ...***Failed    1.14 sec
      Start  6: test_fcl_frontlist
 6/13 Test  #6: test_fcl_frontlist ................   Passed   24.58 sec
      Start  7: test_fcl_math
 7/13 Test  #7: test_fcl_math .....................***Failed    0.01 sec
      Start  8: test_fcl_sphere_capsule
 8/13 Test  #8: test_fcl_sphere_capsule ...........   Passed    0.02 sec
      Start  9: test_fcl_capsule_capsule
 9/13 Test  #9: test_fcl_capsule_capsule ..........   Passed    0.01 sec
      Start 10: test_fcl_simple
10/13 Test #10: test_fcl_simple ...................***Failed    0.10 sec
      Start 11: test_fcl_capsule_box_1
11/13 Test #11: test_fcl_capsule_box_1 ............   Passed    0.01 sec
      Start 12: test_fcl_capsule_box_2
12/13 Test #12: test_fcl_capsule_box_2 ............***Failed    0.01 sec
      Start 13: test_fcl_octomap
13/13 Test #13: test_fcl_octomap ..................***Failed  142.21 sec

62% tests passed, 5 tests failed out of 13

Total Test time (real) = 908.54 sec

The following tests FAILED:
      5 - test_fcl_shape_mesh_consistency (Failed)
      7 - test_fcl_math (Failed)
     10 - test_fcl_simple (Failed)
     12 - test_fcl_capsule_box_2 (Failed)
     13 - test_fcl_octomap (Failed)
Errors while running CTest

One problem is that macro BOOST_CHECK fails without providing useful information. You should consider using BOOST_CHECK_CLOSE instead.

Another problem is that the accuracy in the computation of the closest points is very bad and only works with one gjk solver (GJKSolver_indep if I remember correctly).

It seems that the tests are very unstable.

[panjia1983]

I did not mean the latest commit has no failed tests...What I mean is if you checkout commit on Jan 07( 20147fb1820ddcddb9404616d341d11fbf875979a128), and manually apply the fix in commit 293-295, then all the test should be passed (except one, which was introduced by commit on Jan 7).

I did not have time to fix the bugs introduced by the following commits.

I agree the test is sensitive, since most of them are numerical tests. Will try to replace BOOST_CHECK by BOOST_CHECK_CLOSE.

In addition, I indeed observed libccd has lower accuracy, but we have no control for that library. We can only suggest the user which solver is better.

[florent-lamiraux]

I did, I still have 5 tests that do not pass.
For more efficiency, next time, please create the commit yourself in a temporary branch so that we are sure we talk about the exact same code.

Concerning this pull request, I only reedited another pull request by Karsten1987 were the real patch was hidden in re-indented files. This pull request only adds distance computation between capsules. It should not modify the rest of the code. Feel free to revert part of the pull request.

However, Another more recent pull request (#37) modifies this one for better performance and correct computation of the closest point.

[panjia1983]

Ok. I have only one unpassed (test_fcl_capsule_box_2), though I did not check the octomap stuff.

I am on ubuntu 64bit with gcc 4.6.4.