fit point cloud of object rotate around y axis in superquadric

[jimingda]

Hi,

Recently i test icub-gazebo-grasping, and try to grasp a steer object, some times the test failed and get report "No good grasp candidates found!".

I don't know whether it's because the steer is rotate by y axis about 15 degree, that the superquadric didn't fit well, then the grasp algorithm didn't get good candidates.

As the notes said, current model is supposed to be lying on a table parallel to the x-y plane (to keep things simple). Can i fit point cloud of object that rotate around y axis through some modification?

Thanks.

[pattacini]

Hi @jimingda

As the notes said, current model is supposed to be lying on a table parallel to the x-y plane (to keep things simple). Can i fit point cloud of object that rotate around y axis through some modification?

You can apply a convenient mapping (i.e., a homogeneous transformation) earlier to the input data and the inverse mapping right after the fitting.

However, I don't think this is the point here. The SQ fitting as provided in that repo is clearly not a good option in this case. You don't have to recreate a model from the point cloud¹ as you know already it.

The problem you should address is rather the 3D pose estimation of known objects.

You may want to check out DOPE, for example.

Footnotes

1. https://github.com/robotology/find-superquadric just deals with very simple geometries, whereas the steering wheel is definitely more complex. ↩