Code for the paper RecoveryChaining: Learning Local Recovery Policies for Robust Manipulation (currently under review).
This repository is designed to be used with the Anaconda package manager and has been tested with Python 3.10.
First, install Anaconda or Miniconda. Then, create a new conda environment with the following command:
conda env create -f environment.yml
This will create a new conda environment named recovery_chaining with all the required dependencies.
Next, activate the environment:
conda activate recovery_chaining
Finally, install the package in editable mode. This allows you to make changes to the code and have them reflected immediately without needing to reinstall the package.
pip install -e .
python scripts/learn_recoveries.py failures.learn=True env=pick_place rl=rc_pick_place failures.nfails=100
This will roll out the nominal controllers for the pick-place environment and record 100 failures. The results will be saved in the belief_srs/outputs by default. Move the generated files to a data folder (e.g. data/pick_place) so that they can be reused to learn recoveries.
scripts/learn_recoveries.py recoveries.learn=True env=pick_place rl=rc_pick_place pick_place_data_dir=data/pick_place
This will train a recovery policy to solve the collected failures using RecoveryChaining.
Run the following script to use Lazy RecoveryChaining:
scripts/learn_recoveries.py recoveries.learn=True env=pick_place rl=rc_pick_place pick_place_data_dir=data/pick_place rl.use_nominal_precond=True