takuyahiraoka/efficient-srgc-rl-with-a-high-rr-and-regularization

Source files to replicate experiments in my Arxiv 2023 paper.

What is this?

This is the source code to replicate the experiments provided in ``Efficient Sparse-Reward Goal-Conditioned Reinforcement Learning with a High Replay Ratio and Regularization''.

Demo video

USAGE

REDQ:

python main.py -info redq -env FetchPickAndPlace-v1 -seed 0 -gpu_id 0 -method redq -her -additional_goals 0 -stretch 5

REDQ+HER:

python main.py -info redq-her -env FetchPickAndPlace-v1 -seed 0 -gpu_id 0 -method redq -her -additional_goals 1 -stretch 5 

REDQ+BQ:

python main.py -info redq-bq -env FetchPickAndPlace-v1 -seed 0 -gpu_id 0 -method redq -her -additional_goals 0 -boundq -stretch 5

REDQ+HER+BQ:

python main.py -info redq-her-bq -env FetchPickAndPlace-v1 -seed 0 -gpu_id 0 -method redq -her -additional_goals 1 -boundq -stretch 5

Reset(9):

python main.py -info reset-9 -env FetchPickAndPlace-v1 -seed 0 -gpu_id 0 -method sac -her -additional_goals 0 -stretch 5 -reset_interval 30000

Reset(9)+HER:

python main.py -info reset-9-her -env FetchPickAndPlace-v1 -seed 0 -gpu_id 0 -method sac -her -additional_goals 1 -stretch 5 -reset_interval 30000

Reset(9)+BQ:

python main.py -info reset-9-bq -env FetchPickAndPlace-v1 -seed 0 -gpu_id 0 -method sac -her -additional_goals 0 -stretch 5 -reset_interval 30000 -boundq

Reset(9)+HER+BQ:

python main.py -info reset-9-her-bq -env FetchPickAndPlace-v1 -seed 0 -gpu_id 0 -method sac -her -additional_goals 1 -stretch 5 -reset_interval 30000 -boundq

Note

The main part of this source code is based on the code in [1]. Part of this source code (./customexperiencereplays) is based on the code in [2].

[1] https://github.com/watchernyu/REDQ

[2] https://github.com/ymd-h/cpprb