Towards Safe Navigation Through Crowded Dynamic Environments

Training code for the CNN control policy proposed in our paper "Towards Safe Navigation Through Crowded Dynamic Environments"online, published in 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS).

Requirements:

• Pytorch 1.7.1

Usage:

Assuming you have already collected the dataset and placed it in your home directory. A potential dataset can be used for training is our Semantic2D dataset.

git clone https://github.com/TempleRAIL/cnn_nav.git
# training:
cd cnn_nav
sh run_train_eval.sh ~/dataset/train ~/dataset/dev 
# evaluation:
cd cnn_nav
sh run_eval.sh ~/dataset/test 

Citation

@inproceedings{xie2021towards,
  title={Towards safe navigation through crowded dynamic environments},
  author={Xie, Zhanteng and Xin, Pujie and Dames, Philip},
  booktitle={2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  pages={4934--4940},
  year={2021},
  organization={IEEE}
}