I believe this system was constructed in 2008/2009, which is impressive given the nascent state of open-source drones at the time.
However, today, ArduPilot and Paparazzi are both fairly full-featured in terms of supporting ground stations, waypoints/missions, and autoland, MultiWii/PARIS is very cheap and stable for multirotors (albeit with a more limited feature set), and OpenPilot is developing rapidly. All are open-source and they all use real sensor fusion and stabilization code rather than some sort of Kalman-filter free special sauce like this guy.
It's truly amazing how far we've come in just a few years - ten years ago a small quadcopter would have been hard to imagine (due to a lack of decent micro-sized IMUs as well as a lack of easily-obtainable large high-density LiPos), five years ago a project for only the brave, talented, and well-backed, and now we can buy them as toys!
If you are interested in this, I strongly encourage you to watch Andrew Tridgell's talk from linux.conf.au 2012: https://www.youtube.com/watch?v=ML__e_ZcWiQ. It was close to my favourite talk at that conference.
From the youtube video: "This talk is all about the technology and techniques that we have developed for building an Unmanned Aerial Vehicle (UAV) that can search for and find a lost bushwalker in outback Australia"
Tridge is a great speaker, and discusses the nuances of this topic really well.
> Instead of the usual Kalman filtering, I've devised my own IMU sensor fusion algorithm which is super easy to visualize and doesn't require a PHD in mathematics. Further, it works as well as the complex filters you'd find in the conventional approach.
I wonder what evidence he has for that. It sounds awfully suspect.
In the few cases I've come across where people have claimed this, the stuff behind the curtain has usually turned out to be a muddled simplification of the original made possible by additional constraints.
Think Kirchhoff's circuit laws vs Faraday's law but with less math and more crackpottery.
Which worsens its image further--"I've got an amazing new method better than the scientists could come up with, but it's secret!" smells of incompetence at best.
Lay off the guy. The fact is that the Kalman filter applies to a very small (arguably non-existent in the real world) set of estimation problems, and the solutions that even the PhDs come up with are typically ad hoc and suboptimal.
That being said, there's really not much to his solution - he's relying on gravity (from accelerometer) and magnetometer data to provide most of the orientation information. The IMU really isn't doing much, and I can't see how this works well during maneuvers.
EDIT: Note that's the _actual_ Kalman filter. Most of the time when people say KF, they really mean Extended Kalman Filter, which is an approximation to the real thing. Optimality goes out the window when you approximate.
He's a 3D game engine programmer (the tech cofounder at High Impact Games, an offshoot of Insomnia games - he worked on Ratchet and Clank). He's probably not too bad at 3D programming.
He's not claiming anything big. Just that it works as well as conventional filters (and he might just mean "works well enough").
Sure, he's probably doing something goofy, but it is demonstrably good enough for a toy UAV. Also, I don't think he's claiming a general solution, just something that works for his problem.
Can anybody explain to me how he can legally use amateur radio bands to communicate with this drone? He is a licensed ham, but the device is neither a beacon nor a repeater (and both would need a callsign, at least where I live).
FCC 97.215 [0] makes it legal for a ham to transmit up to 1W in any band they're licensed for for the purpose of model aircraft control, without any callsign identification being necessary.
I reckon legislation will soon fall which makes these illegal unless you are law enforcement or military. If it makes either party potentially more accountable then it's going to cause problems.
However, today, ArduPilot and Paparazzi are both fairly full-featured in terms of supporting ground stations, waypoints/missions, and autoland, MultiWii/PARIS is very cheap and stable for multirotors (albeit with a more limited feature set), and OpenPilot is developing rapidly. All are open-source and they all use real sensor fusion and stabilization code rather than some sort of Kalman-filter free special sauce like this guy.
It's truly amazing how far we've come in just a few years - ten years ago a small quadcopter would have been hard to imagine (due to a lack of decent micro-sized IMUs as well as a lack of easily-obtainable large high-density LiPos), five years ago a project for only the brave, talented, and well-backed, and now we can buy them as toys!