This repository contains the implementations of our paper Grounding Hindsight Instructions in Multi-Goal Reinforcement Learning for Robotics. Both proposed methods HEIR and HIPSS are part of this source code.

@article{Roder_GroundingHindsight_2022,
  title = {Grounding {{Hindsight Instructions}} in {{Multi-Goal Reinforcement Learning}} for {{Robotics}}},
  author = {R{\"o}der, Frank and Eppe, Manfred and Wermter, Stefan},
  journal = {arXiv preprint arXiv:2204.04308 [cs]},
  year = {2022},
}

• git clone https://github.com/frankroeder/hipss.git
• pip users: pip install -r requirements.txt
• conda users: conda env create --file conda_env.yaml

To reproduce the results of our paper, please have a look at the script train.sh

python train.py n_epochs=20 agent=LCSAC env_name=PandaNLReach2-v0

python demo.py --demo-path .\model\LCSAC_20-31-18\
python demo.py --wandb-url <wandb URI: entity/project/runs/trialid>

• Copy example.pyproject.toml to pyproject.toml and adjust the values.
• Install yapf for formatting and pyright for type-checking etc.