Hierarchical Visual Foresight (HVF)

This directory contains code for the paper "Hierarchical Foresight: Self-Supervised Learning of Long-Horizon Tasks via Visual Subgoal Generation" Suraj Nair, Chelsea Finn

Usage

From the google_research directory, run:

virtualenv -p python3.6 hvf

source hvf/bin/activate

pip install -r hierarchical_foresight/requirements.txt

Generate Data

Run python -m hierarchical_foresight.generate_data --savepath=SAVEPATH

Train SV2P

Train a video prediction model using the open source tensor2tensor library. Run git clone https://github.com/tensorflow/tensor2tensor and follow instructions under "Adding a Dataset".

Once you have a trained model on your problem, modify hierarchical_foresight/env/subgoal_env.py to use your model/problem.

Train VAE, TDM, TAP

Train the conditional variation autoencoder python -m hierarchical_foresight.train_vae --datapath=DATAPATH --savedir=SAVEDIR

Train the temporal distance cost function python -m hierarchical_foresight.train_tdm --datapath=DATAPATH --savedir=SAVEDIR

Train the time agnostic prediction baseline python -m hierarchical_foresight.train_tap --datapath=DATAPATH --savedir=SAVEDIR

Run HVF

python -m hierarchical_foresight.meta_cem --difficulty=m --cost=pixel --numsg=1 --horizon=50 --gt_goals=1 --phorizon=15 --envtype=maze --vaedir=VAEDIR --tdmdir=TDMDIR --tapdir=TAPDIR