Skip to main content

MMOneSphere: BC_PCL_PNet2_avg_ee

Here we are now backpropagating through an averaging layer, so this is using segmentation PN++ but regressing to a single EE via an average at the end (just 3D position change). Surprisingly this seems to work very well. In fact these learning curves look REALY good.
Created on April 20|Last edited on April 25
(0.6559 + 0.6214 + 0.4735) / 3 = 0.5836

Train/Eval MSEs, Episode Success/Reward


50100150200Step0.050.060.070.080.09
050100150200Step0.10.150.20.25
050100150200Step00.20.40.60.81
Run set
3


Example GIFs

First seed after 250 epochs. Nice recovery in the bottom right. :)

Second seed after 250 epochs. In some cases the ball went under the water, it's hard to control that.

Third seed after 250 epochs. Looks good. Nice recovery again. :)