Slashdot Mirror
DARPA Developing the Ultimate Auto-Pilot Software
coondoggie (973519) writes "Call it the ultimate auto-pilot — an automated system that can help take care of all phases of aircraft flight-even perhaps helping pilots overcome system failures in-flight. The Defense Advanced Research Projects Agency (DARPA) will in May detail a new program called Aircrew Labor In-Cockpit Automation System (ALIAS) that would build upon what the agency called the considerable advances that have been made in aircraft automation systems over the past 50 years, as well as the advances made in remotely piloted aircraft automation, to help reduce pilot workload, augment mission performance and improve aircraft safety."
This is one of the most often repeated misunderstandings in aviation: the vast majority of crashes is caused by pilots, so we should replace them with automation since that's much more reliable. Errr... no, not by a long shot.
The vast majority of crashes is due to pilot error because the vast majority of possible crashes due to equipment failures are prevented by the pilots. I am a pilot, have never been in a crash, but have had several autopilot and other failures where, if we had not intervened, the aircraft would have crashed. But of course, all those possible crashes due to equipment failures don't make it into the statistics because no actual crash occurred. It's merely a note in the company's safety magazine for crews (along with dozens of others each month). So when an aircraft does crash (even if it's due to equipment failure), it's usually still considered the pilots' fault, and correctly so, because they should have been able to prevent it.
Take the Turkish Airlines that crashed in Amsterdam. Due to a radio altimeter failure during an automatic approach, the aircraft thought it was directly above the runway and pulled the throttles back, while in fact it was still several hundred feet above the ground. Most crews would have seen the speed decreasing (and indeed, this kind of incident had happened many times before to other crews without causing a crash) but this crew reacted much too late and "caused" the airplane to crash.
Or take the Air France that crashed after the pitot tubes froze up. The automation actually failed so the pilots had to take over. Without pilots, the airplane would have crashed anyway. And here, too, this kind of incident had already happened to other crews multiple times, but each time the crew had handled the situation correctly (even though it was not something that was trained in the simulator or accurately described in the procedures). This time the crew did not handle it correctly, in part because they were confused by conflicting warning messages from the airplane's systems telling them the plane was overspeeding and stalling at the same time. They even got aural warnings when they started to, temporarily, apply the exact correction they needed to meke. The automation was not helping them, but actually working against them and telling them they were wrong when they were, in fact, right.
If you want to have an idea of how reliable automation is, just look at the number of military drones that have crashed so far. Their mission couldn't be simpler: take off, fly over some area, come back and land. They only fly in relatively nice weather, there are vaslty less drones than passenger aircraft, yet there are many more drone crashes than passenger aircraft crashes.
It's certainly a good thing that Darpa is trying to make aircraft automation more reliable, but right now pilots are still by far the most important asset for the safety of an airplane.