crazyflie_slam

crazyflie_slam implements a real-time SLAM algorithm for the Crazyflie nanodrone that relies on the FlowDeck for the state estimate and on the RangerDeck for obstacle information.

SLAM

SLAM, which stands for Simultaneous Localization and Mapping, refers to a set of algorithms used in robotics and computer vision that enable a device (such as a robot or drone) to create a map of an unknown environment while simultaneously keeping track of its own location within that environment.

For the Crazyflie there were two main questions to answere before starting.

1. What type of sensors, i.e. shall we use the AI deck with the camera or RangerDeck with 4 Time-of-Flight sensors?
2. Where will the SLAM computations take place? Should we rely solely on the on-board processing power, or should we also utilize an external processor, such as a personal computer, for offloading tasks?

There are two independent choices, resulting in four possible configurations:

• AI deck with on-board processing
• AI deck with off-board processing
• RangerDeck with on-board processing
• RangerDeck with off-board processing ✅

We have opted for the last configuration.

References

We searched for some algorithms that do implement the SLAM task using the available data, i.e. state estimate and four distances. The main resource used is https://github.com/khazit/CrazySLAM

Architecture

The structure of our code is the described in the following image.

As we can see we have a FreeRTOS task added in the Crazyflie firmware that iteratively collect the useful data and then send it over the appchannel to the corresponding endpoint that uses Python-CFlib to read the packet. Furthermore we have the SLAM algorithm that continuously comsumes the received data and updates the model.

Repository structure

• async/: it contains the code for separating the data collection (csv file) and the data consumption for the SLAM algorithm
• crazyslam/: it contains the SLAM algorithm implementation
  • mapping.py: 2d occupancy grid map using ToF sensors range inputs
  • localization.py: state estimate of the Crazyflie using a particle filter
  • slam.py: SLAM algorithm that combines mapping and localization
• imgs
• crazyflie_slam.pdf: presentazione
• my_app_slam.c: FreeRTOS task for collecting and sending data over the appchannel
• README.md: recursive :P
• real_time_slam.py: real-time version of the SLAM algorithm, the CFlib "packet_received" callback adds data in a queue and the Main Thread consumes them and run the SLAM algorithm.
• requirements.txt: required python packages

Installation

• Download repository with git clone https://github.com/JustFanta01/crazyflie_slam.git

• Crazyflie FreeRTOS task:

  • Download the crazyflie firmware
    • git clone --recursive https://github.com/bitcraze/crazyflie-firmware.git
  • Create a new app
    • if present, clean the folder examples/app_hello_world/build, duplicate the folder examples/app_hello_world and rename it as examples/app_slam
    • Copy the my_app_slam.c in the folder examples/app_slam/src/
    • Modify the app_slam/src/Kbuild with obj-y += my_app_slam.o
    • in the examples/app_slam folder run make menuconfig
      • App Layer Configuration --> Enable app entry point --> Save & Exit
  • Cross-compile the firmware with make -j$(nproc) inside the the examples/app_slam folder.
  • Flash the new firmware over the Crazyflie Radio with cfloader flash build/cf2.bin stm32-fw -w <radio address> for example -w "radio://0/80/2M/E7E7E7E7E7"

• Python CFlib

  • Create virtual environment and activate it
    • python –m venv cps
    • source cps/bin/activate
  • Install the required python packages with pip install -r requirements.txt
  • Configure correctly the radio address in the code

Run SLAM algorithm

• for the real-time SLAM run: python real_time_slam.py
• for the async version (CSV file):
  • enter the async folder
  • to create a new csv file logging in real-time the crazyflie data use: python slam_logging_appchannel.py. Otherwise, check the async/datasets for examples.
  • to run asynchronously the SLAM algorithm: python sim_slam_csv.py

Contribution

The project was completed cooperatively by all four team members, with everyone participating in all aspects.

CrazyTeam:

• Luca Fantini​
• Simone Rambelli​​
• Alex Vandi​​
• Pierfrancesco Vazzano​