Skip to content

Uncertainty-driven Trajectory Truncation for Data Augmentation in Offline Reinforcement Learning(ECAI 2023)

Notifications You must be signed in to change notification settings

pipixiaqishi1/TATU

Repository files navigation

Overview

This is the repo of the paper: Uncertainty-driven Trajectory Truncation for Data Augmentation in Offline Reinforcement Learning

Dependencies

  • Python 3.7.13
  • MuJoCo 2.3.0
  • Gym 0.24.1
  • D4RL 1.1
  • PyTorch 1.12.0+cu113
  • TensorFlow 2.10.0

Usage

# for tatu+model-based offline RL
bash run_tatu_modelbased.sh
# for tatu+model-free offline RL
bash run_tatu_modelfree.sh

For different mujoco tasks, some hyperparametes may be diffrent. Please see the original paper for detailed hyperparameters.

About

Uncertainty-driven Trajectory Truncation for Data Augmentation in Offline Reinforcement Learning(ECAI 2023)

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published