Skip to content

Latest commit

 

History

History
107 lines (90 loc) · 6.67 KB

README.md

File metadata and controls

107 lines (90 loc) · 6.67 KB
Raw

VIO-GNSS Recorder

Automatisation of SpectacularAI's GNSS + VIO demo. Including some added functionalities to make collecting data easier. The purpose of this software is to alleviate collecting and analyzing large datasets with SpectacularAI's GNSS-VIO Fusion implementation.

If you have access to a 3D printer, you can use the models provided in /3D_print to print a casing which makes it easier to collect datasets.

Necessary hardware1:

  • u-blox C099-F9P board + ANN-MB-00 antenna
  • Luxonis OAK-D camera
  • Laptop running Linux

Take to account that the antenna likely needs a ground plane to work properly. In our setup a thin circular plane of aluminium worked perfectly fine. The antenna can also be mounted on a car for this purpose.

Automatisation

  • The script vio_gnss_recorder.py asks the user to input some basic information like the authentication credentials for a RTK provider + the user's approximate location.
  • The script initializes NTRIP and the sensor fusion pipeline in the background.
  • After user input the script starts to print the relative pose of the OAK-D device:

Before trajectory alignment

  • User should wait for the module to acquire a RTK Fix (which can be observed from the board's yellow LED). Then traversing 20 to 50 meters should make the trajectories align. The output looks something like this after achieving the sensor fusion phase:

After trajectory alignment

  • The dataset is saved in a session<timestamp> folder.

Map

  • A map showing the travelled route is rendered. The route is determined after Sensor Fusion in the pipe which means that the route should attain a satisfactory accuracy even when temporarily travelling in a GPS dead zone.
  • The map indicates the state of the RTK solution (None, Float, Fix) with colored markers. RTK Fix solution usually corresponds to an accuracy of only few centimeters.2
  • TODO: At the moment, the script utilizes the real-time location output of the system to render the map. This means that SLAM won't correct the rendered route 'in the past'. A functionality could be added so that the map renderer can take SLAM's corrections into account.

Example Map
Example map output (tile provided by OpenStreetMap)

Setup

It is recommended to test the aforementioned demonstration because this implementation utilizes its components under the hood.

This software should be setup and ran inside a Python virtual environment to avoid any conflicts.

Virtual environments in Python. 
pip install virtualenv
# Create and navigate into your project folder
python -m venv <env_name>
# To activate
. <env_name>/bin/activate
# To deactivate
deactivate

Start by cloning these repositories.

git clone https://github.com/AaltoVision/vio-gnss-recorder
git clone https://github.com/AaltoVision/sdk-examples

Configure u-blox C099-F9P board

You'll need to configure the board (instructions also in the PDF). Clone AaltoVision's fork of the u-blox-capture repository3.

git clone https://github.com/AaltoVision/u-blox-capture --recurse-submodules

Navigate to your local copy of the repository.

# for temporary configuration
python ubx_configurator.py DEVICE_PATH example/high_precision_gps_only.cfg
# for permament configuration
python ubx_configurator.py DEVICE_PATH -flash example/high_precision_gps_only.cfg

If the board configuration is not flashed permamently, the configuration has to be done prior to running vio_gnss_recorder.py so after every reconnection or reboot. Important: vio_gnss_recorder.py does not configure the board automatically.

After cloning u-blox-capture you'll need to build str2str

cd RTKLIB/app/str2str/gcc/
make

Dependencies

After activating your virtual environment, install all the necessary packages with pip

pip install -r requirements.txt

Modify config.ini

There's a configuration file config.ini in the parent directory. The absolute paths to the cloned repositories must be provided there so that vio_gnss_recorder.py can use them. Also the device path must be disclosed. The path is usually /dev/ttyACM0 but the user can define the path in the case that they have other peripherals in their setup. It should be considered that the number (/dev/ttyACM0) increments between reconnections of the module so if the board needs to be disconnected momentarily the user must either modify the config file or reboot the OS.

Usage

The workflow for using this automatisation:

  1. Record the data: python vio_gnss_recorder.py
    • You might need sudo rights to communicate with OAK-D in your setup. Either modify udev rules accordingly or use
      sudo env "PATH=$PATH" python vio_gnss_recorder.py
    • The board has an yellow LED which indicates the RTK solution's state:
      • Off: None
      • Blinking: Float
      • On: Fix
    • After acquiring a stable RTK Fix, you should move to align the trajectories. The dataset should be collected after the script starts outputting the global position.
    • Ctrl+C to stop the program. The dataset is saved in a session<timestamp> folder. Locational data points for rendering the map are saved in coords.txt.
  2. Render the map: python render_map.py
    • Renders a map of the traversed route using data points parsed from coords.txt.
    • Asks the user to type some kind of an identifier (eg. Otaniemi_14_8_2023) for the .png to be created.

License

This software is published under the MIT License. Refer to the file LICENSE for more information.

Footnotes

  1. https://github.com/SpectacularAI/docs/blob/main/pdf/GNSS-VIO_OAK-D_Python.pdf — © Spectacular AI ↩

  2. https://sciencing.com/difference-between-rtk-fix-rtk-float-12245568.html ↩

  3. https://github.com/AaltoML/u-blox-capture and https://github.com/SpectacularAI/u-blox-capture ↩