Add wrapper for environments to process hierarchically
The functionality should include:
-
(sub-) goal visualization, at least for MuJoCo Environments -
different goal spaces for different layers. -
different goal thresholds for different layers.
In an ideal world, this would be implemented without modifying the original openai gym environments. The subgoal visualization is probably most difficult, but it may work by online-patching the xml files or loading them in the sim during execution.
Edited by Manfred Eppe