Suspect bug in collide function on master and devel branches

[humanoid-path-planner]

Suspect bug in collide function on master and devel branches

[jmirabel]

I will have a look next week.

[jmirabel]

The behavior when EPA fails has changed indeed. When it fails, it used to return a computed distance which is meaningless. Now it doesn't set the value.

There is no way to recover from EPA failure. The distance may be set to NaN but I don't know all the consequences. The current API does not allow to inform the user of a failure.

[jmirabel]

However, I am surprised you get this error for a box-box query. Can you extract the position of the two boxes and their dimensions ?

Anyway, I agree that when you call collide and you can't provide the penetration depth, you should at least state that there is a collision...

[gabrielebndn]

Setting to NaN or to a negative value would allow to signal failure somehow... It would be certainly better than leaving it undefined.
Anyhow, this does not explain why the Pinocchio unit test used to be working but now it doesn't. That unit test contains both positive and negative examples, and they work. Now it seems computeCollision always returns false

[gabrielebndn]

However, I am surprised you get this error for a box-box query. Can you extract the position of the two boxes and their dimensions ?

I'll see what I can do.

Anyway, I agree that when you call collide and you can't provide the penetration depth, you should at least state that there is a collision...

I do not fully understand the implications of what you said. I analysed the code in a purely functional way, but I did not fully understand what each bit does. If you managed to correctly detect the collision regardless the EPA failure, that would be enough for the Pinocchio test to work

[gabrielebndn]

They are both cubes of side 1.
In the first example they are both at the origin, therefore they coincide, while in the third example the first object is at (0.99, 0, 0) and the second object is at the origin, therefore they should collide.
So, the first and the third example should collide, but they don't.

In detail, here are the placements in the four tests:

  R =
1 0 0
0 1 0
0 0 1
  p = 0 0 0

  R =
1 0 0
0 1 0
0 0 1
  p = 0 0 0

-----------------
/home/gabriele/devel/pinocchio/unittest/geom.cpp(107): error: in "geomTest/simple_boxes": check computeCollision(geomModel,geomData,0) == true has failed
  R =
1 0 0
0 1 0
0 0 1
  p = 2 0 0

  R =
1 0 0
0 1 0
0 0 1
  p = 0 0 0

-----------------
  R =
1 0 0
0 1 0
0 0 1
  p = 0.99    0    0

  R =
1 0 0
0 1 0
0 0 1
  p = 0 0 0

-----------------
/home/gabriele/devel/pinocchio/unittest/geom.cpp(129): error: in "geomTest/simple_boxes": check computeCollision(geomModel,geomData,0) == true has failed
  R =
1 0 0
0 1 0
0 0 1
  p = 1.01    0    0

  R =
1 0 0
0 1 0
0 0 1
  p = 0 0 0

-----------------

[jmirabel]

Thank you for the thorough report. it helps a lot.