Pouring, Reduce Tool Experiments
Now we just have 5 such points in the point cloud! (Edit: now 10)
Created on June 10|Last edited on June 15
Comment
Looks quite promising! On 06/09 I ran 4X cluster 1X locally using BC03 settings.
NOTE: two seeds crashed due to SVD error. If we want to increase the # of tool points, we can just take averages / midpoints of the bottom center with the other boxes (the predicted flow vectors will be a bit different each time which should make it more robust to SVD errors, I think).
Compare this with the same settings (6D flow etc) but with the normal tool points? Note: we did use 6D flow here.
TL;DR if smoothing, 0.5312 for the proposed method, and 0.5064 for the baseline, and if you remove the 2 seeds that crashed (which we probably should do) then 0.5063 so it's very close, and we should just use that value.
UPDATE: now doing more with 10, see next points.
Results
6D Flow SVD Pointwise Consistency (5 tool points)
3
6D Flow SVD Pointwise Consistency (normal tool)
5
10 Points in PCL
Interesting, this actually looks just as good? So now I'm wondering why we need all the tool points. :X
Run set
5
Add a comment