DRIFT-DAgger: Dynamic Rank Adjustment in Diffusion Policies for Efficient and Flexible Training

Setup

Create conda environment for different scenarios

conda env create -f environment.yml
conda activate drift-dagger

Download and train a behavioral cloning policy as the expert using this repo
- The dataset can be found at here

Train a policy with DRIFT-DAgger

python training/mavis/mavis_drift_dagger.py

Name		Name	Last commit message	Last commit date
Latest commit History 2 Commits
config		config
expert_policy		expert_policy
mavis_mujoco		mavis_mujoco
nn_utils		nn_utils
training		training
.gitignore		.gitignore
environment.yml		environment.yml
readme.md		readme.md