SeRO: Self-Supervised Reinforcement Learning for Recovery from Out-of-Distribution Situations

Pytorch code for the submission:

SeRO: Self-Supervised Reinforcement Learning for Recovery from Out-of-Distribution Situations, IJCAI 2023

Supplementary material

• Supplementary material

Requirements

• mujoco200 (https://www.roboti.us/)

Generating and activating conda environment

$ conda env create -f env.yml
$ conda activate sero

Training phase

HalfCheetahNormal-v2

$ cd ~/directory/to/repository/
$ . scripts/train_{algo}_half_cheetah_normal.sh

HopperNormal-v2

$ cd ~/directory/to/repository/
$ . scripts/train_{algo}_hopper_normal.sh

Walker2DNormal-v2

$ cd ~/directory/to/repository/
$ . scripts/train_{algo}_walker2d_normal.sh

AntNormal-v2

$ cd ~/directory/to/repository/
$ . scripts/train_{algo}_ant_normal.sh

Retraining phase (should be executed after the training phase)

HalfCheetahOOD-v2

$ cd ~/directory/to/repository/
$ . scripts/train_{algo}_half_cheetah_ood.sh

HopperOOD-v2

$ cd ~/directory/to/repository/
$ . scripts/train_{algo}_hopper_ood.sh

Walker2DOOD-v2

$ cd ~/directory/to/repository/
$ . scripts/train_{algo}_walker2d_ood.sh

AntOOD-v2

$ cd ~/directory/to/repository/
$ . scripts/train_{algo}_ant_ood.sh

Visualize learning curves

$ cd ~/directory/to/repository/log/
$ tensorboard --logdir={env_name}