This branch contains the Python Simulator for Radio-Inertial Localization with support for Pytorch, Keras and ROS. The MATLAB Simulator for the same can be found on the matlab branch of this repository. The requirements for running the simulator are as follows:

1. Python 3.6
2. Tensorflow/Keras or Pytorch
3. PyLayers

If you use this work (paper or code), or draw inspiration from it, please cite the authors as follows:

@inproceedings{Dhanjal_2019,
   title={DeepLocNet: Deep Observation Classification and Ranging Bias Regression for Radio Positioning Systems},
   url={http://dx.doi.org/10.1109/IROS40897.2019.8967767},
   DOI={10.1109/iros40897.2019.8967767},
   booktitle={2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
   publisher={IEEE},
   author={Dhanjal, Sahib Singh and Ghaffari, Maani and Eustice, Ryan M.},
   year={2019},
   month=nov }

Install Anaconda 3 and proceed with installing Pylayers as per their manual. Once installed, copy and paste all the *.ini files from the assets/inis folder into the pylayers_project/ini/ directory created by the Pylayers installation script. The directory defaults to~/pylayers_project/ini/ on Ubuntu. Since pylayers can be tricky to setup, I've attached my anaconda environment in pylayers.yml. Commit 6bbd9c5 worked best for me.

Several different environment maps have been created for testing out the implementation. Some of them are as attached below:

The blue path is the ground truth generated by the random walk algorithm, whereas the cyan one is the localized path. The red square specifies the goal location and the green square the start location. Actual positions of the access points are encoded in black, whereas those localized are encoded in red (only in Fast SLAM). The results shown below for 2D localization are in the office environment.

The images for all results can be found in the /assets/results/2D/ folder.The result for Fast SLAM using Hard Classification is as follows:

As in the previous case, the blue path is the ground truth generated by the random walk algorithm, whereas the cyan one is the localized path. Actual positions of the access points are encoded in black, whereas those localized are encoded in red (only in Fast SLAM). The results shown below for 3D localization are in the defstr environment.

The images for all results can be found in the /assets/results/3D/ folder.The result for Fast SLAM using Hard Classification is as follows: