Quadstar

Targets of the project

1. Build a airworthy quadcopter from scratch which can be controlled with a joystick
2. Learn PID parameters in a simulation
3. Deploy the learned PID parameters and compare the performance to manually set ones
4. Learn a control policy
5. Deploy the control policy and compare the performance to the PID controller

Controller architectures

PID controller

End2end policy

Drivetrain

BLDC-motors

Axisflying C227 1960KV FPV Motor schwarz

BLDC-driver

B-G431B-ESC1

Akkumulator

Information flow system

Main flight computer

The core of the system is a STM32F411RE microcontroller which computes all processes required for the flight controller which includes sensor fusion.

Sensor extension board

X-NUCLEO-IKS4A1

Simulation

Configuration

All volatile parameters on the user side can be controlled from a single file. When a new model is trained the current config is used for training and is saved in the SQLite database models/config.db.

Plot of quadcopter trajectories

The program quadtest.py can be used to create matplotlib-plots of position, velocity and attitude trajectories from simulation:

python3 quadtest.py

Visualize the quadcopter in the browser with flask

Quadlive provides two functionalities.

python3 quadlive.py

Train models with Stable-Baselines3

To train the a quadcopter flight controller on the current configuration run python3 quadtrain.py.

Delopyment

Joystick

To sample joystick values and send them to the main flight computer run python3 joystick.py.

Credit

The PID controller as well as the flight dynamic model inplementation from quadcopter_simcon have been modified for this reposiory. Big thanks to the creator.