Sunday, February 28, 2010

Near loss of UAV Leads to New Airframe

Two weeks ago we launched our UAV on a routine flight over the Colby campus in light winds. After hand launching the Easystar based UAV form the Lacrosse field, I flew the plane up to about 200 feet AGL (above ground level) and switched over to autonomous mode. The plane started gaining altitude in a circle pattern to the mission perimeters of 400 feet AGL. As the UAV gained altitude, the wind picked up and pulled the plane towards I95. At first we thought that this was merely the UAV following its flight path, but as the it flew farther and father away we realized that something was surely wrong and I retook manual control. By this time the plane was merely a speck on the horizon and I had no idea of the planes orientation because of the planes circling patterns. Operating on false instinct I picked a direction and leveled the plane out. Unfortunately, i guess wrong and steered the further away from us. Flustered, I attemped to gain altitude and turn the plane around, but by this time it was to late and the plane dropped bellow the horizon about a mile from the original launch.

Fearing the worst, Dan and I raced to the car and headed in the direction of the crash. Using a tree as a point of reference we drove up a drive way and to our surprise found the Easystar UAV laying a field of snow unharmed by its emotional run from home. Seriously shaken by the near death of our UAV, Dan enlisted the help of the godfather of our UAV's AttoPilot, the much revered Dean Goude. After looking at the metadata from our flight, Dean sent us this email,

I plotted stabilization data from the 6 flights contained in your LOG file, and compared changes in behavior with tuning progression. I didn't realize (maybe forgot) you are flying a super dynamically stable airframe. This explains the inability to tune pitch stabilization, at least with the recommended forward C.G. for this airfame type. Too low pitch P gain (in SET line $5) then pitch won't hold to target (pitch tends to stay level no matter what) and higher P gain (you went from 10 --> 25 between flight #2 and #3) causes gross oscillation. If pitch is positive there's natural presure pushing nose back down, but with symmetrical gains. On the other hand, Atto simply considers pitch error symmetrically wether or not actual pitch is above or below target. In other words, magnitude of elevator deflection depends simply on pitch error and direction of control surface depends on sign of the error.
AttoPilot's 50Hz attitude control isn't required for such stable planes, and in fact the inherent stability of airframe is a strong detriment to good tuning and flights. Atto is wanting (and more than able) to stabilize squirrely planes but an overly stable model (anything with "Easy" in the name probably) is just fighting Atto from doing its job.
Chris McNair and I talked about this at length on the phone today. As you might know Chris has a Masters in aerodynamics... and his insights here were really key to us figuring out how to help you. Our recommendations to progress on tuning:
1) Move the C.G. rearward, and do so enough so that the airframe is noticably less stable in pitch. Another option is to switch airframes to something with much less dynamic stability... planes with little or no dihedral and with wing NOT much above centerline of airframe. Think of planes like E-Flite Funtana (zero wing dihedral, wing vertical placement positiondown low and on the centerline of thrust)
2) Re-start the tuning process, but with certain values "seeded" as follows:
a) Turn down Alt_D from 10 to 5
b) Steer D to 0 or 2 at most, depending on the amount of dihedral. Polyhedral wings (rudder planes like Miss2) require 0 on Steer_D, but since you have some aileron wing version, and I assume it has only di-hedral (not poly) then maybe 2 on Steer_D. Dynamically stable in roll means Steer_D is not needed very much
c) turn all servo D gains to zero.
I have been able to tune airframes such as yours (Miss2 Old Timer type), but my most honest opinion is these airframes are for simple beginner RC pilots to learn flight. They make great FPV platforms in RC mode, or with simple autopilots like Picopilot (which in fact is tailored for motor gliders, and only similar planes), BUT they are not good airframes for 50Hz attitude control.
Don't get me wrong, I am not saying Atto can't control such planes, but not with nose-heavy CG. De-stabilizing the pitch will allow Atto to control pitch much better. I am working with some customers that use powered parachutes, and they absolutel must control altitude via throtte. Once I get this behavior option coded in and available via SET file, it will control the "Easy" style plane altitude much more directly even with typical nose-forward C.G.
Try the suggestions above, and I do think we can make it work for now. When I say move CG more aft I mean by a healthy amount; perhaps (probably) just outside the recommended rear limit by manufacturer.
AttoPilot LLC
Dean Goedde, Manager"

In layman's terms this means that our Easystar was too inherently stable, sending the Attopilot on a form of infinite loop, and that the Attopilot effectively spent all of its time fixing a problem it couldn't fix. Effectively our highly capable autopilot is designed to fly planes that are harder to fly than our Easystar and doesn't handle inherently stable airframes. This failure to stabilize combined with the high winds caused our near catastrophic crash.

So Dan and I went back to the drawing board. We needed an airframe that was inexpensive, crash resilient, had a large payload, less stable than the Easystar and could handle the variable weather of central Maine. After many hours of research, Dan and I settled upon an EPP Manta flying wing. Flying wings were originally developed during the WWII as long range bombers due to their high efficiency and ability to cary a lot of weight. Here is a video of the EPP Manta in action.

Manta EPP from Gary Mortimer on Vimeo.

Ours should be here on Monday and we will start building and testing immediately.

Thursday, February 25, 2010

I-95 Winter fly over

A few weeks ago Foster and I did a UAV mission over I-95. We configured out autopilot to capture a vertical image every 5 seconds. Post-flight we stitched together 9 images from the flight using PTGUI software. Click on the image below and you'll be brought to our Picasa page where you can download the photo. Zoom in and you'll be able to see some roadside details.