An example of automated photomapping:

typical GPS precision plus turbulence gives around 10-15m precision.

This is what you have after importing automatically, without any manual adjustments, IMU corrections included but the major factor is that the camera is already almost vertical and the platform is stable:

Since the protos are taken with roll-stabilised head using Pteryx UAV,

they are ortorectilinear as the shooting angle was at worst a few deg relative to vertical.

FLEXIPILOT log decoder now take camera lens angle and mounting offset angles as a parameters, generating Google Earth file with matching photos within seconds! This is particularly useful
for isolated shots, since overlapping images when mapped in quantity give an impression of 'randomized crowd'.

Another approach: Microsoft ICE and 10min processing after drag & drop:

Unfortunately the image is a little skewed.

The major benefit of stabilised head is that all photos are of acceptable geometry and useful for stitching. However, because the precise moment of taking the photo cannot be known, an important angle and positional error will always be present in the logs. Another factor to include is changing lighting and altitude, inevitable in real-life operation and during turbulent weather (this time 42km/h flight speed and 20km/h wind speed plus strong thermals affecting pitch).

Automated mapping (AerialRobotics LogDecode) applied to a linear map:

Take a look at this crasy simulation. This is an openloop designed for mission testing.
Of course I got virtual permission and the whole airspace was NOTAMed to the ground.
An UAV has 'flown' over Cote d'Azur.

Simulation time around 2h.
In the previous run I saw mission aborts due to exceeded safety distances etc (due to protective config variables from
real-life photomissions).

In short this is a 2h+ mission, over the range of EasyUAV/Easyglider,Cularis but technically barely flyable with Pteryx loaded with a small camera and some 20Ah of lipos.

In practice taking flight permission from there would be impossible (close to one of France's busiest airports).

Log content:
LOG_POINTS=26274

Flight statistics:
Max altitude above takeoff point AGL=206.0[m]
Cumulative climb=205.97[m]
Max distance to takeoff=9.591[km]
Max VLOS distance to takeoff=9.593[km]
Total travel (groundtrip)=67.950[km]
Takeoff: 2010-04-28 16:30:31.799
Last event: 2010-04-28 18:20:00.046
Total log time=6568.50[s]
Total flight time=6567.00[s] 100.0% of log time
Autonomous flight time=0.00[s] 0.0% of flight time
Average groundspeed=37.2[km/h] 10.3[m/s]
Min air temperature=30.24[C]
Max air temperature=31.51[C]

The answer is:

since UAVs are supposed to fly gracefully,

if you pick a platform that is manageable in manual mode (not overly sluggish),

you can fly a machine of arbitrary size.

Also the stall recovery using rudder is intuitive.

IMO positive stability is not strictly required,

but for real-world applications aircraft stability and autopilot action should better be working together.

MultiplexFox (0.26kg, no payload, 20-45min enduirance, 45-55km/h)

Research plane - classified as Barely Flying Machine. -40deg glide slope when unpowered.

Autopilot is the supporting structure... (maybe World's first Intergral Autopilot Design ;-) )

EasyUAV (1.2kg AUW, 0.65kg empty weight, 20min-1h10 endurance, 35-45km/h)

Pteryx UAV (3.2-5kg AUW, 2.6kg empty weight, 20min-1h30 endurance, 40-55km/h)

Fully custom UAV for aerial photography, not available as a kit or in any other form.

(a joint development of AerialRobotics and TriggerComposites, www.pteryx.eu)

All can be hand-launched, however for Pteryx with over 4kg AUW

it is possible to use bungee launch if you want to operate reliably

even under stress.

Testing the new ArduIMU Viewer with FLEXIPILOT from Krzysztof Bosak on Vimeo.

Testing FLEXIPILOT single gyro and ArduIMU V2 viewer from Krzysztof Bosak on Vimeo.

Accel Yaw: there could be only AccelZ,

calculating any kind of yaw

10 points for very high update rate :-)

Starring ArduPilot GroundStation, MillswoodEngineering GroundStation and Flexipilot. Enjoy!

UAV mission simulation using MillswoodEngineering GroundStation and Flexipilot from Krzysztof Bosak on Vimeo.

UAV mission simulation using Ardupilot GroundStation and Flexipilot from Krzysztof Bosak on Vimeo.

T3-4 Photocontest using EasyUAV from brakar on Vimeo.

UAV flight: New Year's Day 2010 Wrocław, Poland from Krzysztof Bosak on Vimeo.

UAV returning from long range photomapping mission from Krzysztof Bosak on Vimeo.

EasyUAV landing pattern from Krzysztof Bosak on Vimeo.

UAV flight in the rain, automatic landing on the airstrip from Krzysztof Bosak on Vimeo.

UAV fully autonomous long range aerial video flight from Krzysztof Bosak on Vimeo.

UAV search&rescue flight with onboard camera, real sound from Krzysztof Bosak on Vimeo.

UAV low altitude flight, autopilot with IMU from Krzysztof Bosak on Vimeo.

T3 Trust Time Trial robotic airplane pattern flying using FLEXIPILOT from Krzysztof Bosak on Vimeo.

UAV autopilot for aerial photography from Krzysztof Bosak on Vimeo.

AerialRobotics EasyUAV Scenario 3 high-altitude video from Krzysztof Bosak on Vimeo.