Sorry for the late update! We have some new developments to share.
We just released AOgmaNeo/PyAOgmaNeo 1.5.0, which contains several new features as well as internal optimizations.
In this new release, the most notable change is the inclusion of a new ImageEncoder. This new ImageEncoder is now itself (optionally) hierarchical, allowing for higher quality image representations. You can get some pretty crisp image reconstructions with this new ImageEncoder!
The Python bindings now have a ton of runtime checks that give (hopefully) useful information to the user. This is good for both debugging and learning how to use the library. If high performance is desired, one should use the C++ version (regular AOgmaNeo), which doesn't have any runtime checks.
Internally we started using more handles to make copying hierarchies less error-prone.
We are also working on some cool new demos, that use our UBL (Unsupervised Behavioral Learning) paradigm instead of regular RL (Reinforcement Learning). You can ready about on older demo that uses UBL here, hopefully UBL will replace RL in the master branch at some point. There is an especially interesting new demo that uses it on the horizon, will reveal more as it gets closer to completion.