BC03 3DoF, 1 Sphere, PN++ SVD Pointwise Flow
WITH the consistency loss. NOTE: cannot look at the evaluation flow visualizations or MSEs because that doesn't have the correct model.eval() but the action selection should be the same. Unfortunately doesn't look that good?
Created on May 13|Last edited on May 16
Comment
With the consistency loss.
Edit: I think we need to look carefully at this pipeline, train/MSE_loss seems to be OK but leads to poor evaluation performance. Also it's a bit higher than if we had no consistency loss but I think that makes sense if the network has to balance two competing priorities.
Flow targets are in the range (-1,1) per component. From training it seems like it is able to accurately tell when to go up and down? Why is sideways movement not predicting anything:
We cannot look at the eval/MSE_loss or eval flow visualizations (these are pre-May 14).
Section 1
Run set
5
Add a comment