bennevans / iida

Context is Everything: Implicit Identification for Dynamics Adaptation

Project Page

Installation Instructions

1. Clone the repo

git clone [e-mail]:bennevans/iida.git
cd iida/

2. Install MuJoCo here.

3. Install the required python packages:

conda env create -f setup/environment.yaml

OR

cd setup/
pip install -r requirements.txt

3. Install the varyingsim library

pip install -e . 

4. Generate the simulated dataset or download it here.

Dataset Generation Instructions

Single-step Environments

cd scripts/dataset/
python create_da_dataset_push_box.py
python create_da_dataset_slide_puck.py

Multi-step Environments

To train a policy to generate rollouts, I use mjrl.

cd scripts/policy/
python train_policy.py
python relabel_paths.py --env <env>

Run instructions

To run with the current experiment and algorithm parameters:

cd scripts/

And run:

python run_exp.py -c <config file>.yaml