ugglan's Introduction

Ugglan

Ugglan (owl in Swedish) is a hobby drone project. Making use of the Raspberry Pi Zero 2 W (amongst other things).

The project serves as a learning platform for the author himself and is quite broad, containing everything from assembler to docker. It also contains some multi body modelling and simulation as well as state space estimation and control.

Setup

Prerequisites

python >=3.9
docker
make

Run the setup script in the root.

source ./setup.sh

Documentation

Sphinx is used for the generation of the documentation.

Build

The documentation is build using make.

make -C ./doc

The generated html documentation is found under ./doc/build.

ugglan's People

Contributors

Stargazers

Watchers

ugglan's Issues

Improve handling of env variables

E.g., create a class that collects all env variables and provides methods such as printing. A lot of double code now e.g., see print_env_vars in main.cpp.

Build dynamic lib of target implementation for analysis env

E.g, instead of the double implementation of the attitude estimation in Python, re-use the target implementation in c++.

Only write imu config if needed

First read and then write if mismatch. Add debug logging.

Refactor c++ catch unit tests

Write streaming client GUI.

Add docker image caching in ci pipeline

Improve gyro offset compensation.

Don't require standstill during start-up e.g., check if standstill.

Update attitude estimation performance figures when happy

Use default factory for all numpy arrays

Add support for my new MacBook M2

Build release dpkg (using qemu platform emulation)

https://github.com/docker/setup-qemu-action

Fix flaky timing tests

Task
DataLogQueue
AfroEsc angular-rate
...

Low-pass filter acceleration & update Kalman noise covariance matrix

Due to vibration's the IMU acceleration needs to be low-pass filtered before entering the Kalman sensor fusion in the attitude estimation. Also, the R matrix should be computed such that the estimation can cope with linear acceleration e.g., dynamic R which favors the gyro more during acceleration (the acceleration should be used for long-term estimation due to bias in gyro).

Send and visualize live data

E.g., send the raw data stream over tcp and visualize in a GUI.

Non-linear 6dof simulation

Implement the full (6dof) non-linear model to enable for more complex offline simulation. Some stuff which needs to be done:

Refactor the attitude estimation, make clear interfaces s.t the estimators can easily be re-used.
Implement the non-linear model. Should this be a class e.g., NonLinearModel(Estimator, CONSTANTS, ...)?
Create a GUI to execute and analyze the simulation in e.g., Tkinter.
Determine values of the non-linear model i.e., drag coefficient.
Need to be able to add noise e.g., sine or gaussian to estimator input.