Bug in the Python version example for attaching a camera

[c-joly]

Hi @costashatz, here is an issue associated to the Joss review https://github.com/openjournals/joss-reviews/issues/6771

I tried to attach a camera to the end of the pendulum link and the result was not as expected. After reviewing the sample code in the FAQ (which is the same than the one in cameras.py), I figured out a bug. Here is the original proposed code:

tf = dartpy.math.Isometry3()
rot =  dartpy.math.AngleAxis(3.14, [1., 0., 0.])
rot = rot.multiply( dartpy.math.AngleAxis(1.57, [0., 0., 1.])).to_rotation_matrix()
tf.set_translation([0., 0., 0.1])
camera.attach_to_body(robot.body_node("iiwa_link_ee"), tf) # cameras are looking towards -z by default

For the C++ version, we have:

Eigen::Isometry3d tf;
tf = Eigen::AngleAxisd(3.14, Eigen::Vector3d{1., 0., 0.});
tf *= Eigen::AngleAxisd(1.57, Eigen::Vector3d{0., 0., 1.});
tf.translation() = Eigen::Vector3d(0., 0., 0.1);
camera->attach_to_body(robot->body_node("iiwa_link_ee"), tf); // cameras are looking towards -z by default

In the C++ version, we can see that tf is initialized with the rotation parameters, which is not the case in the Python version. Then, the Python version created the rot matrix to adjust the orientation of the camera (I think it is to compensate that camera looks towards the -z axis), but it does not use it! Thus, I added the following line in the code between the two last lines, and the result looked as expected:

tf.set_rotation(rot)

Can you confirm my patch is correct and update the example and doc accordingly?

Last question: why do you need to move the camera of 0.1 on the z axis? Is it to be a little bit more far from the node to avoid some kind of artifact if we put the camera exactly on the node?