rc-flight is the high level flight controller for the Rice Creek UAS project. It is distributed under the MIT license except where noted (see the licenses subdirectory for details.) rc-flight is an embedded autopilot application for unmanned aircraft systems.

• Watch rc-flight in action, raw and unedited
• Visit my blog for a variety of rc-flight (formerly AuraUAS) related projects and background information

We know you have a choice in autopilots, so we appreciate you flying with us! The focus here is simplicity and understandability without sacrificing performance and quality. There is a very intentional emphasis on writing as much of the code as possible in pure python. There is a limited number of supported environments and sensors to avoid complicating the architecture and build system.

rc-flight includes:

• An advanced 15-state EKF (extended kalman filter) developed by the University of Minnesota Aerospace Engineering Department. Internally the filter uses quaternion math to avoid gimbal lock issues.

• A sophisticated route management and waypoint following system.

• A circle hold system that holds precise circles even in strong winds.

• A flexible/configurable PID-based control system. Several controller options are available including a classic PID controller and a velocity form controller, along with digital filters and summers. These can be mixed and matched to create multistage control systems. Gains and configurations can be tuned from the ground in real time.

• TECS values are computed and available as inputs to the PID controllers.

• Flexible device driver system.

• Property system: a system for managing and sharing hierarchically organized data between program modules and scripts. The proprety system enables shared state (and communication) between C++ and python modules within a single hybrid application.

• Native python script execution and integration on board the host flight controll (usually a beaglebone for my projects.)

• Flexible actuator support.

• Great circle distance and heading math.

• Self learning IMU temperature and magnetometer calibration system.

• System and health monitoring

rc-flight is built with the python setup.py tools. Under the hood there are several critical C++ modules that are packaged and wrapped for python.

1. Run: "./setup.py build" in the top level directory.

2. Run: "sudo ./setup.py install"

Special note: if you find your pi or beaglebone is getting crushed during the compile and running out of memory, consider setting up a swap partition [temporarily] during the compile. Instructions are in the README-beaglebone.md file. This can really improve your native compile experience if you are building directly on the embedded comptuer. Don't forget to remove the swap partition when you are finished building because it burns a lot of your avialable disk space.

AuraUAS is a complete (and independent) open-source autopilot. It features an inexpensive DIY hardware option and a professionally built (pick and place) hardware option. The design philosophy is built around creating a high quality, robust core set of features using a standardized hardware platform. Hardware was selected to strike a balance between keeping the system inexpensive for a hobbyist budget, while maintaining very high standards of performance and robustness.

This is not a project that attempts to do everything for everyone with every possible sensor and every possible use case. That approach adds tremendous complexity and other projects are pursuing this approach very effectively.

Aircraft are designed and developed with light weight as a priority in every phase. Analogous to that principle, AuraUAS is developed with simplicity as a priority. Aircraft don't always end up as light as we would like and AuraUAS is not as simple to understand as I would like, but know that this is a priority and something we push for at every step.

A core element of our 'simplicity' philosophy is that the autopilot code is divided into two main parts that run on separate processors.

1. All the sensor data collection and PWM servo control and the manual fail safe mode run on a Teensy-3.2 (or 3.6.) The firmware is developed in the arduino enviroment and addresses all the hard real-time needs of the autopilot system using a simple interrupt service routine model.

2. All the higher level functionality runs on an embedded linux platform. Currently this is a beaglebone, but the code can easily run on any linux-based system. The linux code is single threaded and uses non-blocking I/O strategies to maintain real-time 100hz performance.

   As part of the push towards simpler code, the linux-side application is a hybrid mix of C++ and python3. I have found that I can easily maintain 100hz real-time performance, even with a significant amount of python code in the main loop. Opinions may differ, but I find that python leads to 40-50% fewer lines of code, and that code can be much more readable. Embedded scripting also enables powerful feature development without needing to modify the core C++ code and recompile the firmware.

Together, the system is comprised of two simpler applications working together as a distributed system. You can compare this to the typical open-source autopilot architecture that results in a giant monolithic application which uses a complicated (and often brittle) thread-based architecture to maintain near real-time performance.

Over the past few years AuraUAS is evolved into a mature and stable autopilot system. It is a rock solid work horse for many of the University research projects I support. Still, with any system, there are endless feature changes and improvements to be made.

Immediate development goals include:

• recently the fmu firmware was ported from arduino to ardupilot's chibios-based AP_HAL to eliminate the need for custom in-house hardware development. The next step is to integrate all the changes into an airplane and flight validate this big change.

• allow the fmu to optionally run the configurable pid controllers.

• allow the fmu to manage telemetry and logging.

• longer term my hope is to migrate all the C++ specific code directly onto the fmu with enough basic functionaly to support common tasks. The rc-flight code then will act more like a host (or companion) computer for managing higher level mission tasks or doing experimental tasks "offboard".

• [done] Support for structured hdf5 data log export.

• [done] Python 3 port.

• [done] Implement a hand-thrown auto-launch task. This is written as a python task (within the python-based mission system) and produces very very solid and stable launch results for hand thrown airplanes (such as a Skywalker or Talon.)

• [done] Add support for computing survey routes on board the aircraft. (The issue here is the brittleness and time required to send 100's of waypoints up to an aircraft over a radio modem link. Instead we can just send the area to be covered and some camera parameters and the aircraft can compute it's route internally.)

• [done!] Redesign and modernize the inexpensive reference hardware.