How to run `snn4hrl`

Stochastic Neural Networks for Hierarchical Reinforcement Learning (snn4hrl) as presented at ICLR by Carlos Florensa, Yan Duan, Pieter Abbeel (https://openreview.net/forum?id=B1oK8aoxe&noteId=B1oK8aoxe)

To reproduce the results, you should first have rllab and Mujoco v1.31 configured. Then, run the following commands in the root folder of rllab:

git submodule add -f https://github.com/florensacc/snn4hrl.git sandbox/snn4hrl
touch sandbox/__init__.py

Then you can do the following:

Train a SNN for the Swimmer environment via python sandbox/snn4hrl/runs/train_snn.py
Look at the visitation plot including the visitations of every latent code in data/local/egoSwimmer-snn/
Train a hierarchical policy on top of that SNN via python sandbox/snn4hrl/runs/hier-snn-egoSwimmer-gather.py

florensacc/snn4hrl