BC04 (Pouring) Naive PN++ Methods

Created on June 10|Last edited on June 11
Comment
As of 06/10/2022, running a bunch of comparisons of this naive PN++ to vector, comparing with and without scaling of targets.
06/11/2022: good news, naive PN++ looks like it maxes out at 0.48. The main issue is that it is very noisy. :) Compare with our SVD methods for pouring https://wandb.ai/mooey5775/mixed_media/reports/BC04-Pouring-SVD-Flow-Based-Methods--VmlldzoyMTQ5MDMy  we see higher and less noisy performance.
Results﻿
eval/info_done_final
eval/info_done_final
0100200300400500Step00.20.40.60.81
BC04_PourWater_v01_ntrain_0100_PCL_PNet2_eepose_3DoF_ar_8_hor_100_rawPCL_noScaleTarg_2022_06_10_10_10_10_0001   PN++ to Vector (no scale targ)
BC04_PourWater_v01_ntrain_0100_PCL_PNet2_eepose_3DoF_ar_8_hor_100_rawPCL_scaleTarg_2022_06_10_17_38_09_0004   PN++ to Vector (scale targ)
BC04_PourWater_v01_ntrain_0100_PCL_PNet2_eepose_3DoF_ar_8_hor_100_rawPCL_scaleTarg_2022_06_10_17_38_09_0003   PN++ to Vector (scale targ)
BC04_PourWater_v01_ntrain_0100_PCL_PNet2_eepose_3DoF_ar_8_hor_100_rawPCL_scaleTarg_2022_06_10_17_38_09_0002   PN++ to Vector (scale targ)
BC04_PourWater_v01_ntrain_0100_PCL_PNet2_eepose_3DoF_ar_8_hor_100_rawPCL_scaleTarg_2022_06_10_17_43_22_0001   PN++ to Vector (scale targ)
BC04_PourWater_v01_ntrain_0100_PCL_PNet2_eepose_3DoF_ar_8_hor_100_rawPCL_scaleTarg_2022_06_10_17_38_09_0001   PN++ to Vector (scale targ)
PN++ to Vector (no scale targ)1
PN++ to Vector (scale targ)5
﻿
GIFs, no scale target﻿
GIFs, scale target﻿
Add a comment