Having designed and flown R/C sailplanes in the past, it seemed an obvious idea to loft a GPS-guided glider by balloon, and have it return to the launch site.  Usually, balloon / radio groups loft an insulated package with a parachute, and track it down wherever it happens to land.  With our home region surrounded by saltwater, that method wouldn't be wise here in any event. 

After some back-of-the-envelope doodling, it became obvious that the pressure variations, high airspeed, fast jetstream winds, and very low temperatures would make it quite a challenge to get up past 65,000 feet, where the sky is black, and the horizon curved.  A big issue was that the glider's wing loading, and hence landing speed, would need to be as high as possible to be able to fly fast enough at altitude to push back through the jetstream.

The first draft used the traditional landing method of using large flaps to lower the landing speed and steepen the glideslope on the approach.  I soon realized that serious pitch trim issues would be involved which would be beyond all but a very sophisticated autopilot and INS nav system, a system that would be massive overkill for the remainder of the glider's flight.   Plus, even with a moderately steep glideslope, when you take into account all the errors involved, a very large and smooth landing field would be required, especially given the need to clear surrounding trees.  Improbably large, in fact, for a largely forested region.

I had experimented with parachute recovery systems for large R/C aircraft in the past, so this seemed an obvious solution.  Some fairly simple autopilot algorithms could pop the chute the right distance upwind, and a much smaller and rougher field could be used.  A chute would also provide a "bail-out" option if other systems failed to work properly.

Later, while browsing other balloon-launch sites, I learned someone else had the same general idea, years ago.  Right down to the parachute for landing.  Nevertheless, in the better part of a decade of work, they hadn't really made it as far as you might expect.  And, I had far more ambitious altitude goals.

The project seemed an ideal and challenging combination of my existing skills in software development, full scale aviation, navigation, and small sailplane design and construction, and a great opportunity to learn more about a lot of other fields. 

It's turned out to be more than just a small challenge.  New areas included both analog and digital electronics, radio telemetry, low-level packet-networking, "real time" software systems,  feedback and control, high-reliability coding, in-depth work with GPS.  Plus, a fair bit about oscillation and damping, perhaps more than I would have liked to learn. 

Original Physical Spec:

Altitude Goal:  85,000 feet ASL (26 km)

Maximum mass: 2.75 kg

Design Mass Goal: 2.2 kg

Temperature Tolerance, external:  +30c to -70c

Max Temperature Swing, internal:  +50c to -10c

Battery Duration:  3.5 hours flight, 5.0 hours with reserve

Reliability:  Recoverable after failure of any single part, excepting computer.

Indicated Air Speed at Best Glide (sea level):  40 knots (20 m/s)

Best Glide Slope:    (12 : 1) or better

Landing Descent Speed:   < 4 m/s

Landing Field Required:   150 m x 150m or smaller

Abilities / Design Spec

The overarching goal was to use this project as an exercise in developing a complex, real-world system that was reliable, robust, and fail-safe.  The general approach was to take each system and component and say, "What if this failed completely?  What if it shorted out, or became only partially disconnected?", and then ensure there was a way to continue flying, or at a minimum recover the glider intact.

To this end, early on a goal was set of having the glider be almost completely autonomous - once it's launched, it should be capable of flying back completely on its own.  There shouldn't be any crutches such as manual guidance to for the final-approach to landing required, or decision making as to what to do if something goes wrong.  All such functions should all be onboard, but with the ground station still able to intervene and tell the "expert" pilot to take an action, or fly the plane from the ground at the autopilot or manual control level if desired.

Ideally, I also wanted to have the glider work without having to resort to using an "artificial horizon" reference, such as an internal-nav system driven by solid-state gyros and accelerometers, as that would radically increase the cost, risk, and complexity.  It also makes it a better and more challenging project, to see if it can be made to work with the hardware that is actually required, and no more.

The goal would be to build a navigation system with just GPS, pitot / static, and compass as primary nav instruments.  Early on, the idea of including a small desktop web cam was dropped as too much trouble, but was added to the final design to serve as a crude backup "position sensor" for manual-piloting, in the event the GPS failed.  Even fairly low quality images from a webcam, at a rate of 1 per minute, would be enough to locate the glider and steer it in to visual range by compass / autopilot, and this left only a single component - the onboard computer itself - as a critical path.

In the end I came very close to this goal, eventually requiring only one low-cost gyro, used as a roll rate sensor, to allow the autopilot to fly the aircraft in a robust way.