UBCMOCCA/SymmetricRL

Does adding artificial phase (/time) help?

Closed this issue · 1 comments

farzadab commented

The Phase_Walker2DBulletEnv-v0 has an added artificial phase variable which might be useful to the other algorithms as well, so to be fair we might need to compare the others with phase as well.

farzadab commented

Experiments on Walker2D

Based on the following plot I don't think we need to be worried about that, though we can do the comparison to be more thorough.
image

Experiment IDs:

  • 2019_07_08__23_53_20__walker-fixstd-lossmirror
  • 2019_07_08__15_22_11__walker-fixstd-netmirror
  • 2019_07_08__15_22_02__walker-fixstd-nomirror
  • 2019_07_10__14_57_55__walker-fixstd-phasemirro
  • 2019_07_08__15_21_50__walker-fixstd-tupmirror

BTW, the walker can still get much better with more training, maybe we should increase the limit (same plot):
image