The LAPAC algorithm was developed as a course project for COMP 767 -- Probabilistic Graphical Models -- at McGill University, given by Professor Siamak Ravanbakhsh in the Fall 2019 semester.
LAPAC is a maximum entropy reinforcement learning algorithm that is designed particularly for partially-observable environments. It uses a fully stochastic latent variable model inspired by SLAC, and uses the generative latent dynamics model to train the actor and critic offline.
The project report can be viewed at this link.