/dal

Repository for Deep Active Localization research and benchmarks

Primary LanguagePythonMIT LicenseMIT

Deep Active Localization (DAL)

Repository for Deep Active Localization research and benchmarks. Accepted to RAL. https://ieeexplore.ieee.org/abstract/document/8784238, https://arxiv.org/abs/1903.01669

Requirements:

  • Python 3.5+
  • Pytorch 1.0
  • OpenAI Gym
  • Numpy
  • tensorboardX

Please use this bibtex if you want to cite this repository in your publications:

@article{gottipati2019deep,
 title={Deep Active Localization},
 author={Gottipati, Sai Krishna and Seo, Keehong and Bhatt, Dhaivat and Mai, Vincent and Murthy, Krishna and Paull, Liam},
 journal={IEEE Robotics and Automation Letters},
 volume={4},
 number={4},
 pages={4394--4401},
 year={2019},
 publisher={IEEE}
}

Installation

Clone this repository and install the dependencies with pip3:

git clone https://github.com/montrealrobotics/dal
cd dal
pip3 install -e .

Basic Usage

For running our gym environment (which we call dal-v0), you can just do python main.py (More instructions will come soon but the code and parameters used are mostly self explanatory. you can also look at a2c_ppo_acktr/arguments.py) For training or testing on our custom simulator, see: sim/readme.md

Acknowledgements

  • Gazebo
  • pytorch
  • openAI gym
  • Ikostrikov's baselines repo
  • BabyAI