Data labeling

[andreaprotopapa]

we need to get from the acquired markers the observation vector → Data Labeling

• the state space S is given by S=[p_t, p_g, p_c], so we need to identify which marker is part of which object and its relative position inside the object

  • p_t = X positional 3D key points of the trunk (along its medial axis)
  • these are computed as average from the cable markers
  • p_g= 3D position of the goal point
  • p_c = 3D position and orientation of the cube

• to identify the markers, see this documentation on Data Recording and Trained Markersets

  • with Trained Markersets we can track objects of any kind of shape. They are set of points trained to be recognized under a cusom flexible skeleton given as input

    Trained Markersets allow you to create Assets from any object that is not a Rigid Body or a pre-defined Skeleton. This allows you to track anything from a jump rope, to a dog, to a flag, to anything in between.

• how to export data from optitrack

  • see this documentation on Data Export

Actual Marker setting

[andreaprotopapa]

MARKER LABELING UPDATES:

• [x] rigid tracking of each ring through asymmetric markers (s.t. we distinguish each ring)
• [ ] implement our tracking method —> wuth hungerian matching algorithm or optical flow
• [ ] pointcloud —> through chamfer distance or dexpoint

[andreaprotopapa]