This repositroy contrains the code of the paper Symmetric Models for Visual Force Policy Learning
- Install BulletArm v1.1 (unreleased)
- Clone this repository
git clone https://github.com/ColinKohler/SymmetricVisualForceLearning.git
- Install dependencies from Pipfile
pipenv install
- Create results directory to store train logs and model
mkdir SymmetricVisualForceLearning/data/
python scripts/train.py [task]
Replace [task] with the desired domain to train on. Currently avaliable tasks are:
- Block Picking: block_picking
- Block Pulling: block_pulling
- Block Pushing: block_pushing
- Block Pulling Corner: block_pulling_corner
- Mug Picking: mug_picking
- Household Picking: clutter_picking
- Drawer Opening: drawer_opening
- Drawer Closing: drawer_closing
- Peg Insertion: peg_insertion
For example, to run block picking:
python scripts/train.py block_picking
For additional information, you can examine the training script arguments in the normal manner:
python scripts/train.py --h
The simulation configurations and hyperparameters can be found in configs/task.py.