Greetings!
I'm currently working on Pick and Place using Kinect camera, detecting a box from the PointCloud.
As it's shown in the pictures below, by using several PCL methods the box-like cluster is getting successfully detected (as the biggest cluster on the table).
In my understanding, the goal of the pipeline is **to publish the result as a TF transform** for the arm to move towards and grasp.
Unfortunately, while trying to broadcast a transform, the origin sets up just in the corner of the box, but not in the middle/top, which is not good for moveit_commander to work with.
Following [Source 1](http://wiki.ros.org/tf/Tutorials/Writing%20a%20tf%20broadcaster%20%28C%2B%2B%29) and [Source 2](https://industrial-training-master.readthedocs.io/en/latest/_source/session5/Building-a-Perception-Pipeline.html), tf Broadcaster seems to be set as so:
static tf::TransformBroadcaster br;
tf::Transform part_transform;
// Here in the tf::Vector3(x,y,z) x,y, and z should be calculated based on the pointcloud filtering results
part_transform.setOrigin( tf::Vector3(sor_cloud_filtered->at(0).x, sor_cloud_filtered->at(0).y, sor_cloud_filtered->at(0).z) );
tf::Quaternion q;
q.setRPY(0, 0, 0);
part_transform.setRotation(q);
// Broadcast a transform
br.sendTransform(tf::StampedTransform(part_transform, ros::Time::now(), world_frame, "part"));
Am I mistaken at some point?
Have someone faced the same problem of *"tf in the corner"*?
Feel free to ask and clarify any moments!
Best Regards,
Artemii
↧