PyRTKLIB -- A bridge between AI and GNSS.

News

2025.10.02

We roughly adapt our binding to rtklib_demo5 2.5.0 EX. But it's not fully tested. Please kindly report issue when you have trouble. And we have a more pythonic frontend called TASGNSS. Feel free to have try.

Our paper has been successfully published in IEEE Transactions on Intelligent Transportation Systems! If you find our tool useful, please cite our paper, thanks so much!

@ARTICLE{10965937,
  author={Hu, Runzhi and Xu, Penghui and Zhong, Yihan and Wen, Weisong},
  journal={IEEE Transactions on Intelligent Transportation Systems}, 
  title={pyrtklib: An Open-Source Package for Tightly Coupled Deep Learning and GNSS Integration for Positioning in Urban Canyons}, 
  year={2025},
  volume={26},
  number={7},
  pages={10652-10662},
  keywords={Global navigation satellite system;Deep learning;Python;Artificial intelligence;Weight measurement;Satellites;Receivers;Mathematical models;Training;Position measurement;Artificial intelligence;deep learning;GNSS;RTKLIB},
  doi={10.1109/TITS.2025.3552691}}

2024.10.25

1. We opensourced a new light urban dataset with LOS/NLOS label and other sensor data. Please find and play in KLTDatset.

2. We optimized the file structure, and put the example script to the example folder to avoid error importing issue. The example observation data is also uploaded to the folder.

2024.09.23

The preprint version of our paper pyrtklib: An open-source package for tightly coupled deep learning and GNSS integration for positioning in urban canyons is now available on arxiv.

We would greatly appreciate it if you could cite our work:

@misc{hu2024pyrtklibopensourcepackagetightly,
      title={pyrtklib: An open-source package for tightly coupled deep learning and GNSS integration for positioning in urban canyons}, 
      author={Runzhi Hu and Penghui Xu and Yihan Zhong and Weisong Wen},
      year={2024},
      eprint={2409.12996},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2409.12996}, 
}

2024.09.22 v0.2.7 Experimental Features - Tightly coupled deep learning and GNSS integration

1. A tightly coupled deep learning and GNSS integration subsystem is currently under development and several useful functions are implemented(e.g. weight least squares in rtk_util.py)! For more details, please refer to the dev repo TDL-GNSS.

2. A File Wrapper has been introduced to manage the context of file descriptors. In previous versions, the file handler lacked proper context management. In this version, the "FILE*" parameters are replaced by a FileWrapper. For more details, refer to the definition in cbind.h. Additionally, the previous expansion const char *filename, const char *mode is still supported.

   Here is an example: Suppose you need to use an RTCM stream from an NTRIP server, and the file may be updated frequently. We can use the following program to create a FileWrapper:

   import pyrtklib as prl
   
   def get_rtcmf(rtcm, fp):
       ret = 1
       while ret != -2:
           ret = prl.input_rtcm3f(rtcm, fp)
       fp.cleareof()  # Clears EOF to continue reading
       return rtcm
   
   rtcm = prl.rtcm_t()
   prl.init_rtcm(rtcm)
   eph_file = prl.FileWrapper("eph.rtcm3", "rb")
   get_rtcmf(rtcm, eph_file)
   

   Note that the cleareof() method uses clearerr() to reset the EOF status of file reading, allowing the program to continue reading from where it left off rather than starting from the beginning.

2024.07.10 Add windows support

The release packages now are built with github actions. And windows support is testing. They are available on linux, macos, and windows now. If you find any bug, please submit issues.

2024.07.08 v0.2.6 Bug fix

• In cbind.h, I forgot to allocate one more byte for string to store '\0', thus segmentation fault may occur in function "convertChar" (I met several times). Besides, in convertType, the copy size is not correct and only the first object is copied. All the bugs now have been fixed.

2023.11.23 v0.2.5 Optimize the folder structure

In previous versions, the .so file is directly in site-packages folder, which is disgusting. To let the code editor can make use of the .pyi, the .so file is moved, and is loaded by the init.py file now.

2023.11.07 v0.2.4 Update the rtklib to demo5 b34h

Update the base rtklib version to demo5 b34h

• The support number of Beidou up to 46
• No implementation functions are deleted. I don't know why they remain in the source code.
  • input_tersusf
  • input_cnavf
  • readfcb
  • init_cmr
  • free_cmr
  • update_cmr
  • input_cnav
  • input_tersus
• And these files are deleted
  • delete pyrtklib5/rtksrc/ppp_corr.c
  • delete pyrtklib5/rtksrc/qzslex.c
  • delete pyrtklib5/rtksrc/rcv/cmr.c
  • delete pyrtklib5/rtksrc/rcv/gw10.c
  • delete pyrtklib5/rtksrc/rcv/rcvlex.c

2023.11.05 v0.2.3 Bug fix and code optimization

This version contains follow updates:

1. Using template binding functions to replace binding macros.
2. Arr1Dchar now has an own constructor and can be constructed from a python string.

   output_path = Arr1Dchar("pyexample_output.txt")
   

3. Now all the pointer member in structure has a setter function, you can directly modify it like:

   obs = obs_t()
   data = Arr1Dobsd_t(100)
   obs.data = data
   obs.n = len(data)
   obs.nmax = obs.n
   

   But it's not recommended to do so, because this may lead memory leakage. Many pointers in rtklib may refer to anther variable, thus, call free() to directly clear it may cause other problems. Please make sure you won't modify it too often.
4. obs.set_data() has been removed.
5. "FILE*" params in functions (input_ubxf, input_rawf e.g.) is devided into "const char *filename, const char *mode", the file open operation will be processed in the overloaded function. Here is an example.

   raw = raw_t()
   ret = input_rawt(raw,STRFMT_UBX,Arr1Dchar("example.ubx"),Arr1Dchar("rb"))
   

6. Fix a serious bug on MacOS in rtkcmn.c. On MacOS (or BSD), _POSIX_C_SOURCE should larger than 199309, or the function strtok_r will have a different behavior, and make segmentation fault.
7. Optimized the example.

2023.10.14 Install via pip

Big Progress!!!!

Now you can install it via pip

pip install pyrtklib5

Introduction

This is a Python binding for RTKLIB from rtklibexplorer , which provides more satellites support compared to the original version. Many researchers are currently using Python for research, especially in deep learning field. Thus, we implement this Python interface of RTKLIB to build a bridge between Python and positioning. By means of RTKLIB, you can easily read data from rinex file and process the positioning using the methods provided by RTKLIB, such as SPP, RTK, PPP.

If you want to use the tool based on original RTKLIB version, please refer to pyrtklib

If you want to use the rtklib deeply, please refer to the point position example.

If you just need the result, please refer to the post position example.

Installation

Install Via Pip (Recommend)

pip install pyrtklib5

The package relies on both system version and python version. The release packages are built with github actions. They are available on linux, macos, and windows now.

Manually Install

Installation of pyrtklib is very easy, just clone the code and install it.

0. Dependencies

   We don't provide pre-compiled package so far, thus, this library will be compiled in time. (Looking forward one day it can be directly installed by pip!) It requires several dependencies, including gcc and cmake. If you are using on a MacOS, please first install gcc by brew and set the compiler path, because RTKLIB used GNU C features, which is hard to solve in clang. MSVC is not tested.

1. Clone from github

git clone git@github.com:IPNL-POLYU/pyrtklib_demo5.git

2. Install

cd pyrtklib_demo5
python3 setup.py install

Then you can use this lib by importing it.

from pyrtklib5 import *

Key Feature

The aim for this binding is to make a interface of RTKLIB, without changing any code in RTKLIB. However the biggest obstucle is that RTKLIB enoumously uses function side effect. For example, the reading function readrnx, you should pass the obs, nav, sta as input parameters, and the process result will be directly written to obs, nav, and sta. Thus, these functions only receive the pointer and it requires the Python interface contains the original pointer. Unfortunately, the STL container can't satisfy this requirement because STL container will copy the data. If Python receives a STL container, it's not the original version of the data.

Thus, we implement Arr1D and Arr2D to handle the array and can also be used as the pointer. Arr1D has a structure like this:

template<class T>
struct Arr1D<T>{
    T* src;
    int len;
}

If you want to create it in Python, just use "Arr1D"+"ctype", for example, if you want a char type array, then, it should be:

myarr = Arr1Dchar(3)

The constructor parameter is the length. In the above example, the length is 5. You can directly to modify the data:

myarr[0]='a'
myarr[1]='b'
myarr[2]='c'

And you can also use the slides:

yourarr=myarr[0:2]

So far, the Arr1D and Arr2D don't have the ability to print itself, if you use

print(myarr)

you'll get:

<pyrtklib.Arr1Dchar object at 0x7fe3799a6ef0>

So, if you want to see the content, you can change it to a list by:

list(myarr)

Then the result will be:

['a','b','c']

<span style="color:red;font-size:20px">Warning: Be really carefull to use Arr1D and Arr2D, because it doesn't have a bounary check. In some cases, it's a pure pointer and doesn't know how long itself is. So be careful to set the index, or the segmentation fault will occur.

Usage

This binding provides all the functions and data structures in rtklib. The structures in rtklib is used as a class in Python. For example, if you want a obs_t, then you can use it by:

obs = obs_t()

And the data in obs can be accessed by:

obs.data[0]

Notice that obs.data is a Arr1Dobsd_t class. In other words, this is an array of obsd_t. For more detail, you can see in the example.py.

Citation

If you find this tool useful, you can cite our paper as:

@ARTICLE{10965937,
  author={Hu, Runzhi and Xu, Penghui and Zhong, Yihan and Wen, Weisong},
  journal={IEEE Transactions on Intelligent Transportation Systems}, 
  title={pyrtklib: An Open-Source Package for Tightly Coupled Deep Learning and GNSS Integration for Positioning in Urban Canyons}, 
  year={2025},
  volume={26},
  number={7},
  pages={10652-10662},
  keywords={Global navigation satellite system;Deep learning;Python;Artificial intelligence;Weight measurement;Satellites;Receivers;Mathematical models;Training;Position measurement;Artificial intelligence;deep learning;GNSS;RTKLIB},
  doi={10.1109/TITS.2025.3552691}}

@inproceedings{hu2023fisheye,
  title={Fisheye camera aided NLOS exclusion and learning-based pseudorange correction},
  author={Hu, Runzhi and Wen, Weisong and Hsu, Li-ta},
  booktitle={2023 IEEE 26th International Conference on Intelligent Transportation Systems (ITSC)},
  pages={1--8},
  year={2023},
  organization={IEEE}
}