Planning and Execution using Inaccurate Models with Provable Guarantees

BibTeX Citation

@INPROCEEDINGS{Vemula-RSS-20, 
    AUTHOR    = {Anirudh Vemula AND Yash Oza AND J. Bagnell AND Maxim Likhachev}, 
    TITLE     = {{Planning and Execution using Inaccurate Models with Provable Guarantees}}, 
    BOOKTITLE = {Proceedings of Robotics: Science and Systems}, 
    YEAR      = {2020}, 
    ADDRESS   = {Corvalis, Oregon, USA}, 
    MONTH     = {July}, 
    DOI       = {10.15607/RSS.2020.XVI.001} 
}

Dependencies

Most of the dependencies are listed in the requirements.txt file, and can be installed using the command

pip install -r requirements.txt

Make sure to install the gym package that is provided locally instead of the latest version to ensure compatibility

pip install -e gym

To run the 7D PR2 experiments, we need additional dependencies given in the external/ folder that can be installed as follows

pip install -e external/pyglet
pip install -e external/pyopengl
pip install -e external/urdfpy

Reproducing Experiments

Simulated 4D Planar Pushing

To run the experiments corresponding to the fetchpush environment (Table I in paper and Fig 2 (right)) run the following command inside src/ folder

python -m odium.experiment.experiment_fetch --exp-agent <agent> --exp-model <model>

where <agent> should be one of {'rts', 'dqn', 'mbpo', 'mbpo_knn', 'rts_correct'} where

rts corresponds to our approach,
dqn corresponds to Q-learning
mbpo corresponds to Model learning with neural network model (NN)
mbpo_knn corresponds to Model learning with K-nearest neighbors model (KNN)
rts_correct corresponds to always planning with accurate model,

and <model> should be one of {'accurate', 'inaccurate'}.

Contributors

The repository is maintained and developed by Anirudh Vemula from the Search based Planning Lab (SBPL) at CMU.