Pytorch implementation of Conservative reinforcement learning algoritm for random-delay environments.
Paper link: Conservative RL, Neural Networks 2026
python == 3.8.0
pytorch == 2.0.0
mujoco == 2.2.0
mujoco_py == 2.1.2.14
gym == 0.26.2
python main.py --env-name HalfCheetah-v3 --min-obs-delayed-steps 0 --max-obs-delayed-steps 10 --init-obs-delayed-steps 10 --delay-type uniform --random-seed 2026 max-step 1000000
@article{lee2026reinforcement,
title={Reinforcement Learning via Conservative Agent for Environments with Random Delays},
author={Lee, Jongsoo and Kim, Jangwon and Jeong, Jiseok and Han, Soohee},
journal={Neural Networks},
pages={108645},
year={2026},
publisher={Elsevier}
}