crazyflie-stepStabilizer

Semesterproject for lecture Systems-on-chip for Data Analytics and Machine Learning.

Height estimation with a Neural Network
The general approach of enabling drones to perform altitude estimation in indoor environments relies mainly on a fusion between inertial information and distance measurements to the ground. Usually, the latter provides more precise information, performing measurements relative to the ground with centimeter-precision - using the time-of-flight (ToF) principle. However, while very precise, using this type of sensor on drones comes with a limitation. If the floor is not perfectly flat, or if there are various obstacles on the floor (e.g., tables, chairs), the drone will experience a sudden change in altitude when hovering above these obstacles. This sudden change can destabilize the drone or can have an undesired effect on flight behavior. The goal of this project is to detect these effects using not only the distance raw data but also other onboard sensors (IMU, barometer) and predict the height with an NN. Thus in this project, you will: 1): Collect data on the sensor data and ground truth of height to form a dataset. 2): Train a simple neural network to give good height estimation 3): Port that neural network over to the MCU we have onboard (STM32). Note that this step can be done in multiple ways, TFLite/Cube.AI/CMSIS-NN, etc.

Installation

Clone the repo and change directory:

git clone [email protected]:Xeratec/crazyflie-stepStabilizer.git
cd crazyflie-stepStabilizer

Install dependencies:

$> chmod +x configure
$> ./configure

Clean and build project:

$> make clean
$> make

Make targets

all        : Shortcut for build
compile    : Compile cflie.hex. WARNING: Do NOT update version.c
build      : Update version.c and compile cflie.elf/hex
clean_o    : Clean only the Objects files, keep the executables (ie .elf, .hex)
clean      : Clean every compiled files
mrproper   : Clean every compiled files and the classical editors backup files

cload      : If the crazyflie-clients-python is placed on the same directory level and
             the Crazyradio/Crazyradio PA is inserted it will try to flash the firmware
             using the wireless bootloader.
flash      : Flash .elf using OpenOCD
halt       : Halt the target using OpenOCD
reset      : Reset the target using OpenOCD
openocd    : Launch OpenOCD

Flashing

Writing a new binary to the Crazyflie is called flashing (writing it to the flash memory). This page describes how to flash from the command line and there are a few different ways to do it.

Using Crazyradio

The most common way to flash is probably to use the Crazyradio.

Prerequisites

• A Crazyradio with drivers installed
• crazyflie-clients-python placed on the same directory level in the file tree
• The firmware has been built
• The current working directory is the root of the frazyflie-firmware project

Manually entering bootloader mode

• Turn the Crazyflie off
• Start the Crazyflie in bootloader mode by pressing the power button for 3 seconds. Both the blue LEDs will blink.
• In your terminal, run make cload

It will try to find a Crazyflie in bootloader mode and flash the binary to it.

Warning: if multiple Crazyflies within range are in bootloader mode the result is unpredictable. This method is not suitable in classroom situation where it is likely that several students are flashing at the same time. Also remember that the Crazyradio PA often reaches into the next room.

Automatically enter bootloader mode

• Add the address of the crazyflie to the config.mk file, for instance CLOAD_CMDS = -w radio://0/80/2M
• Make sure the Crazyflie is on
• In your terminal, run make cload

It will connect to the Crazyflie with the specified address, put it in bootloader mode and flash the binary. This method is suitable for classroom situations.

Note: this method does not work if the Crazyflie does not start, for instance if the current flashed binary is corrupt. You will have to fall back to manually entering bootloader mode.

TF Lite

Because all of the firmware for the crazyflie is written in C, and TF-Micro is in C++, we need to compile TF Micro with some limited C++-11 support and standard libraries for math.

Therefore, the TF Micro is compiled in C++ with some important functions exposed in C and then later linked with the crazyflie firmware.

Authors

Philip Wiese (ETHZ ETIT)
[email protected] - Xeratec

Luca Rufer (ETHZ ETIT)
[email protected] - LucaRufer

License

This software is released under Apache 2.0, see the LICENSE file in the root of this repository for details.