Comment by Earl on August 24, 2012 at 6:06pm

We are doing this also, but with an X8 airframe. So far only got to 50 feet then had a premature release.
What autopilot did you use? I hope the RTL of APM2 works. What was the problem? GPS quit or AP quit? We flew 6 balloons to almost 100,000 ft and recovered science package every time. We also use redundant APRS tx. Did you get any telemetry? We always did. Using 3dr 433mhz radios with great success. Now we have a 5 watt ATV live video tx in the nose to see in real time to fly by mavlink and to be able to change modes. We are in Abq,Nm so come see us.
Earl

Comment by Harry on August 24, 2012 at 6:26pm

WOW!  What altitude was it in the video @8:30 or so when it stopped spinning and went straight and level?  What prevented straight and level before that? 

This stuff is amazing, so much I need to understand before I could try it.

Comment by Greg Fletcher on August 24, 2012 at 6:59pm

Wow! That was crazy. I wondered how the imu would hold up from a HAB drop. It never could get into a stable solution due to all the high velocity thrashing around and using these accelerations to correct gyro drift? It never had a chance. And the thumping sound and visual oscillations(camera shaking) I think were your servos flopping back and forth as fast as they could.

When you say the "new AHRS code" do you mean the one from 2.6?

You may need different gain schedules for different altitudes and speeds. Maybe a quick and dirty auto tune algorithm to quell the cycling? If I was doing a HAB drop, heres what I would do.

1)Use a releasable drog chute to keep more stable after drop.

2)Modify the AP code with the ability to switch from thermopile sensors to the AHRS depending on which one makes  sense. I have an ancient set of sensors from pre-gyro days. They are very reliable and never get mixed up. And no, you can't have them, I may need them for a project. ;) Close to space there are no clouds or hills to mess them up. They sense the only the horizon, and I think work better up there. APM 1 & 2 have the analog pins brought out, so just a little soldering.

Any how great try. It's the first attempt I've seen. Very cool!

Comment by Larry Grater on August 24, 2012 at 7:44pm

Make sure to check out the quicklook .ppt.  I plotted a bunch of the data and have comments supporting some of the conclusions.  In reality this may take a little time to work out.  I actually complted scaling code for HIL and flight, but didn't have a representative model in X-Plane, then FG so there was no way to really evaluate goodness past what's in the baseline code.  This can be done several ways, but here Qbar scaling on rate and airspeed on PI is probably right, but needs some code mods to implement. The most troubling issues will be with the 2.50 imu compensation.  I'll be digging in here once I've finished the rest of the HW evalutations.  So far nothing looks like it failed.  One aileron horn is broken + aileron attachments were fairly loose.  The could have been crash damage, the later is from all that limit cycling.  I'm amazed that the airframe is this robust.

The altitude that it went S&L was about 15-17k ft.  Look at the plots.  The airspeed scaling went from 31 to 25 m/s here.  Part of what's happening is that the AS was reading low on Qbar and V and it was impossible to get to the command.  The other part is that the vehicle was pulling 4 g's in a coning motion and having issues turning the velocity vector at this speed.  The lower than actual Qbar/V measurement creates a mismatch in the stock scaling and contributes to the limit cycling in both axes.  Not quite sure why we blasted through the -40 deg limit on theta, but this section is not stock and I'll go back and look. 

Larry G

Comment by Greg Fletcher on August 24, 2012 at 8:54pm

The first ArduPilot worked with infrared optical sensors. It can do it again, with gyros and such.

Comment by Andrew Rabbitt on August 25, 2012 at 3:54am

Wow Larry!  Congratulations.  There's a bunch of info and lessons for everyone to be gleaned from your work here, so thanks for posting about it.

I have been away working in Germany at the moment, so my HAB glider efforts are at a standstill and I am not fully up-to-date with where the APM code is going at the moment, so obviously my comments below are based on my work earlier and older code.

From my limited simulation experience then, it looks like you definitely have an altitude/density related gain problem there.  In addition, I have never got my simulations to work well straight after the first drop with any form of closed-loop control/navigation running.  The 'aircraft' always tumbles and APM can never recover on its own.

The way I have successfully simulated is to drop initially in manual mode, no control inputs.  Once reasonably above stall speed I switch to stablize (wing leveler) mode.  After the gravity vector is roughly pointing in the correct direction, I then switch to full RTL or Auto mode.  This I do all manually, but obviously it eventually needs to be coded and automated.  I would bet that even your Stinger airframe when trimmed correctly would fly in manual pretty well at 89000' although probably initially with a depressing glide ratio.

Anothing thing I would point out is that the vortex shedding from the aerial would possibly disturb the aircraft vertical stabiliser as well as adding a random pitch/roll/yaw signal to the IMU.  I would consider repositioning it to the leading edge of the vertical stabliser or embedding it in a wing or in some way eliminate its aerodynamic effect.

Regarding the gains, I don't know how detailed you went into the code, but there are some clips on the speed correction that caused me problems with my simulation early on.  They're fine for conventional flying but way too limiting for high altitude flight.  I'm happy to help you work though all the gain scaling etc if you like.  Unfortunately I'm not in a position to run any X-plane or FG simulations at the moment.

After all that, I still have to say WOW and thanks for sharing the results of bleeding edge high altitude flight! 

 

Moderator Comment by Gary Mortimer on August 25, 2012 at 4:23am

@Carl and @greg the space shuttle used thermopiles as well ;-)

Comment by Harry on August 25, 2012 at 6:00am

This stuff really is cool.  I read the patent for FMA's copilot years ago and there was mention of space craft in there.

I see now that my questions were answered in the first post, got all excited by the awesome content so sorry to ask the obvious.

Comment by Harry on August 25, 2012 at 7:04am

Here's a paper which is about a thermopile autopilot which isnt news to anyone here, but the NASA technical documents it references are about orienting spacecraft using thermopiles.  The FMA patent is referenced too.  Just thought I'd link it in case someone might find it useful. 

http://www.ctie.monash.edu.au/hargrave/horizon_sensing_autopilot.pdf

Comment by Larry Grater on August 25, 2012 at 8:01am

Thanks for posting link to the paper. 

This morning I'm working a bit on video analysis.  The 1st tasks are to time tag critical frames, determine roll rate time history during descent.  I think the atttiude initial condition is solvable by looking at the sun angle.  In the mean time I have a couple of pictures that I'd like to share.  Note that we were not so clear of the balloon debris.

Comment

• ‹ Previous
• 1
• 2
• 3
• Next ›
• Page