BC04 (Pouring) SVD Flow-Based Methods
This should be for our 'proposed' 3D and 6D flow methods; ablations can go elsewhere.
Created on June 10|Last edited on July 15
Comment
TimelineResults (CoRL Supplement + Beyond)Results (Pre-CoRL 2022 Deadline)GIFs3D Flow (Scaling PCL/Flow)3D Flow (no scaling PCL/Flow)6D Flow (Scaling PCL/Flow)6D Flow (no scaling PCL/Flow)
Timeline
(06/10/2022) Getting this started and comparing our proposed SVD methods (both 3D and 6D flow), with and without scaling of the flow / PCL.
(06/11/2022) Looking promising, doing some more runs. Also it looks like results here are better than the naive way https://wandb.ai/mooey5775/mixed_media/reports/BC04-Pouring-Naive-PN-Methods--VmlldzoyMTQ4OTYy :)
(06/18/2022) Running additional experiments due to the supplement.
Results (CoRL Supplement + Beyond)
Here's me trying to group the 3D flow methods, with consistency lambda values: (0.0, 0.1, 0.5, 1.0), with 5X runs each.
Edit 07/15/2022: unfortunately I can't figure out how to group the runs. By "group runs" I mean I want their performance curves averaged to form 1 curve that I see in the plot below for eval/info_done_final along with the standard errors shaded in. I report this in this wandb post here: https://community.wandb.ai/t/how-to-group-runs-e-g-different-random-seeds-together-on-the-wandb-report-function-for-plots/2634
3D Flow, PM, NO Consist.
5
3D Flow, PM, Consist 0.1
5
3D Flow, PM, Consist 0.5
5
3D Flow, PM, Consist 1.0
5
Results (Pre-CoRL 2022 Deadline)
3D Flow SVD PW Cons, Scaled
5
3D Flow SVD PW Cons, RAW
1
6D Flow SVD PW Cons, Scaled
5
GIFs
3D Flow (Scaling PCL/Flow)
3D Flow (no scaling PCL/Flow)
6D Flow (Scaling PCL/Flow)
6D Flow (no scaling PCL/Flow)
Add a comment