Mistmach between GJKSolver::shapeIntersect and GJKSolver::shapeDistance

[jcarpent]

From the current code version, it appears that the methods GJKSolver::shapeIntersect:

https://github.com/humanoid-path-planner/hpp-fcl/blob/ea9afdc2ee5771826a12a769a108368cd0c57212/include/hpp/fcl/narrowphase/narrowphase.h#L57-L61

and GJKSolver::shapeDistance:
https://github.com/humanoid-path-planner/hpp-fcl/blob/ea9afdc2ee5771826a12a769a108368cd0c57212/include/hpp/fcl/narrowphase/narrowphase.h#L212-L216

returns opposite outputs when there are collisions.
Is it expected?

[jcarpent]

@jmirabel @florent-lamiraux Maybe you have some hints on this issue?

[jmirabel]

Can you be more precise on what is opposite ? I cannot guess it from the code.

[jcarpent]

What I meant is that shapeIntersect returns True if there is a collision while shapeDistance will return False and vice-versa.

[jmirabel]

gjk.shapeIntersect() == true naturally means there is a collision. This one should not be changed.

gjk.shapeDistance(distance) == true conventionally means distance > some_security_threshold. Personally, I find it natural that it returns true in this case (personal taste) and I don't see any objectively good reason to change this convention. (The behavior being different from shapeIntersect is clearly not strong enough for me).

If you want the same interface, you can add a method like shapeCollisionFree or shapePenetration (I don't advise the latter).

[jcarpent]

Looks clear to me. I will add shapeCollisionFree in the near Future.