What a mess. We botched GPS/AHRS integration in our 2005 Grand Challenge vehicle. Seen this type of problem.<p>Here's the problem. You have an Attitude and Heading Reference System (AHRS) with three rate gyros and three accelerometers, plus a magnetometer as a compass. From this information, you want to get aircraft orientation and heading. Integrating the rate gyros gives orientation, but because you're integrating a rate, there's cumulative error over time. The accelerometers give you a down reference. Not always a good one. Some AHRS units will lose their down reference if you fly in a circle for a while, and the accelerometers see a consistent but wrong "down" direction from centrifugal force.<p>So AHRS systems are prone to cumulative error accumulation. You also have a GPS system, which is prone to short-term noise but does not accumulate cumulative error. GPS is position only; there's no orientation info from GPS. (There are some multiple-antenna GPS systems that get orientation, but those are rare) So AHRS/GPS systems try to combine the two using various filters. Those filters embody assumptions about the error properties of each system. Fusion of the two sources provides position information, with the GPS fixes being augmented with short-term info from the AHRS. However, it's possible for GPS position to "jump" due to radio propagation problems. You see this on smartphones all the time. It's not as bad for aircraft, which usually have a clear view of the sky. (It's much harder for ground vehicles, which can't always see enough GPS satellites. We had a lot of trouble with this in 2005.)<p>The important outputs are attitude (pitch, roll, and yaw) which drive the artificial horizon ball (or, today, a graphic which looks like one). There's also heading, which drives the compass. The attitude outputs also drive the aircraft's stability control systems. The AHRS data can be used raw, or combined with GPS data to get not only attitude but position.<p>Usually, the AHRS outputs are used without GPS info to drive the aircraft systems that just need attitude. But the fused AHRS/GPS data is usually better, and dealing with inconstencies between the unaugmented AHRS info and the fused data is a pain. So there's a temptation to use the fused data for everything. Embraer apparently did this.<p>There's an argument for source integration. Air France 447 crashed because they lost airspeed and altitude data when the pitot tubes and static ports froze up. An integrated AHRS/GPS system would have given them accurate info on vertical speed and altitude even without air sensors. So integration isn't fundamentally a bad idea. If you integrate sources, though, you have to deal with sensor conflicts. That's a hard problem. Filters alone will not do it.<p>The integration used by Embraer (who did the avionics, by the way?) apparently doesn't handle GPS failures properly. The AHRS is still there, and that's all you need for aircraft stability. But if the fused outputs are used for stability augmentation, stall warning, and such, apparently they can fail when the GPS data does.