nicehiro / RIS

Goal-Conditioned Reinforcement Learning with Imagined Subgoals

Environments

The multiworld environments are taken from https://github.com/vitchyr/multiworld/tree/leap

Ant Navigation

To train RIS on the U-shaped ant maze environment, please run:

python train_ant.py --env_name AntU

Use this table to run RIS on other ant maze navigation tasks:

Environment	--env_name
U-shaped ant maze (default)	AntU
S-shaped ant maze	AntFb
$\Pi$-shaped ant maze	AntMaze
$\omega$-shaped ant maze	AntFg

Vision-Based Robotic Manipulation

To train RIS on the vision-based robotic manipulation environment, please run:

python train_sawyer.py