Jack Crossfire's blog View Details
Archive for November, 2007 - Page 2
Posted by Jack Crossfire | Nov 11, 2007 @ 04:17 AM | 2,964 Views
Lousy weather is expected for 2 weeks, so it's programming time.

So now we have a waypoint recall/store switch. Vicacopter stores a single waypoint on the flash. Also have the operator mode switch, switching the right stick between horizontal and vertial/heading in operator directed mode. Finally, the camera switch, which disables the automatic shutter to save the Aiptek flash.

Engaging Recall mode before or after engaging autopilot causes Vicacopter to return to the stored waypoint when autopilot is engaged. In Store mode, she does not return to the stored waypoint when autopilot is on. In both modes, the stored waypoint is updated whenever an operator direction is given.
Posted by Jack Crossfire | Nov 10, 2007 @ 06:10 AM | 2,902 Views
Well, as U probably know, the NTE Xena diodes at Fry's suck. They only read their rated voltage for a single current.

Fortunately 4 us, the ADXRS150's have 2.5V reference taps. Didn't want to use those originally because it requires another long wire & header. The SPI users R really ahead of us now. They have all these voltage issues covered inside the gyros, which brings to mind Vicacopter II.

http://www.centuryheli.com/products/...?currentid=191

The answer is yes, U can get 600 size semi symmetric blades and that brings a Trex 600 back into range. Hopefully aerial photography will increase production of semi symmetric blades, but as dysfunctional as U. Know. Where. is, it probably won't happen unless it's Google approved.

Unfortunately the Micrmag3 has failed completely. It booted once after the last rework. Then it stopped booting. There is no autopilot without the Micromag3.
Posted by Jack Crossfire | Nov 09, 2007 @ 01:51 PM | 3,404 Views
So the once perfect gyros R now showing significant drift. The voltage regulator appears unstable. Battery level, number of peripherals, voltage regulator temperature, RF signals all may change the gyro bias. It changes by 0.07V in a single flight, even more than the IDG300's.

Unfortunately, no HD aerial this week. Just too many problems with the gyros and with the dot com boom, commute times R back to 2 hours.
Posted by Jack Crossfire | Nov 08, 2007 @ 02:14 PM | 2,909 Views
In addition to the LED shifting voltages by 0.03V, looks like the gyros pick up a 72Mhz RF boost when the remote control is nearby. The transmitter previously had to be on to record calibration data.

Now we're keeping the transmitter off during calibration and recording data at all times instead of just when the transmitter is on.

So much for that. The next bug was when transmitter contact was lost & regained, the flight recording was flushed, locking up the CPU momentarily. A lot of delayed neural network cycles then got run in succession, locking out further flight control.

Finally, the rebuild had significantly more vibration. The blades slipped out of track again and the gyros were not immune to it as hoped. So frequent blade retracking is required.

There are 2 filters in the gyro pipeline. An averaging filter on the PIC creates aliases of vibrations over 25 Hz. Then a lowpass filter on the ARM rejects vibration over 13Hz. This scheme seems to pass through more vibration when the blades are tracked because the tracked blades have different vibration modes. Need to move all the lowpass filtering to the PIC.
Posted by Jack Crossfire | Nov 07, 2007 @ 01:38 PM | 2,997 Views
So rebuilt the electronics table for the 4 wheel camera mount and autopilot stopped working. After the rebuild, the status LED started interfering with the analog conversions. Clearly when the LED is on, Vmax is 0.03V lower than when it is off. Never happened before.

What we need to do is power the LED off the servo bus instead of the CPU bus, 1 day of work. The increased noise in the pitch gyro is another matter.
Posted by Jack Crossfire | Nov 06, 2007 @ 02:46 AM | 2,961 Views
Onboard camera view during daylight autopilot (0 min 41 sec)


Finally, we have the onboard photos recorded during the daytime autopilot demo. According to the flight recording, Vicacopter spent 14 minutes in auto hover, the longest ever. The Aiptek couldn't compress it as well, so it only stored 50 photos.
Posted by Jack Crossfire | Nov 05, 2007 @ 10:52 AM | 3,275 Views
Well, our bid to photograph the space station didn't go so well. Got a big boxy dot and a small round dot trailing it. They definitely orbit real low. Should probably get a real telescope for these bright objects.

As for the following bid to get aerial, timelapse footage of a sunrise, forget it. The Aiptek just doesn't have enough memory. 64 photos aren't enough. That only gives U 10 minutes. At least the auto hover was stable. Also, we need waypoint following. Just can't fly over 150ft above eye level without it.

At least we got some shots of auto hover at eye level.
Posted by Jack Crossfire | Nov 04, 2007 @ 11:45 PM | 2,997 Views
So we now have the neural network hard wired to match tilt directions with acceleration directions. That enabled 260% more iterations, possibly better altitude prediction, but no obvious breakthrough. At least in the current super stable air, she's hovering like a pro at 154ft & 11 satellites.

The formula is:

Inputs:
prev accel
current tilt

Intermediate:
6 nodes

Outputs:
predicted accel

The problem with the 10" long straps in the pontoon camera mount is they're not as strong as shorter straps. Now 4 another idea.
Posted by Jack Crossfire | Nov 03, 2007 @ 07:58 PM | 3,024 Views
Got the attitude blending by interpolating Eulers and to no surprise, she hovered facing South and did an operator directed piroette with ease. In fact, the accelerometer results in a euler interpolation need only 1/3 the weight of the accelerometer results in a quaternion interpolation to keep the gyros under control. Until the next time yaw estimation breaks.

At this point in Vicacopter's modeling career, she's very stable during photo shoots. The inertial navigation is working so well, it's like the South facing problems never were. We have a good idea of her wind limits. We've characterized her requirements for GPS acquisition & stable orientation acquisition. Time to retire the 6in -> 3out neural network and implement 2in -> 1out.

Had a miraculously calm afternoon so it was time to try the Aiptek in daylight. On attempt 1, the Aiptek didn't boot at all. On attempt 2, the Aiptek recorded 21 pictures but its flash erased itself for some reason. Attempt 3 was devoted to ground pictures with the 200mm so it didn't have much altitude. There it got 32 pictures and didn't lose the flash.

The Aiptek may erase its flash if another camera is plugged into the USB bus simultaneously.

Now for a movie of all 32 surviving Aiptek photos in auto hover.

The Aiptek view in auto hover (0 min 32 sec)


Don't forget, on Mon Nov 05 05:54 AM what's left of the space station will be flying over dumpy apartment from the SW to the NE....Continue Reading
Posted by Jack Crossfire | Nov 03, 2007 @ 01:53 AM | 3,153 Views
Could have sworn this mount was already tried, but like a dot com boom, it's back for a 2nd engagement. We strap the upper half of the foam tubes to the skids and the lower half to the camera platform. The key to all the best mounts was transferring the lift force through loose foam.

The last test flights for the 6in -> 3out neural network are trying to debug the yaw problem. It's a miracle of artificial intelligence and inefficiency that this neural network can autonomously correlate tilt angles & throttle to acceleration.

Unfortunately it's so addicting to hit autopilot in the stable North heading and stand under her LED glow, we don't get much flight data in the unstable South heading.
Posted by Jack Crossfire | Nov 02, 2007 @ 12:18 PM | 2,787 Views
So much for laptop replacements. Should have known those $350 Walmart sales were meaningless. They sold out 5 minutes before they even opened. The worst part of those sales is getting up early and going through all that trouble for something U never have a chance of getting. They're like TV game shows. Hate being treated like a monkey...or an American.

So the biggest problem is rising prices. Those laptops were $420 in August, then $500 in September, and now $600. It would be worth paying $600 now and not using it, just to avoid future inflation, but the batteries only have 1 year of storage life.

Now for possible vibration suppression mounts for Vicacopter. The most successful mounts have no hard connection between camera & copter.
Posted by Jack Crossfire | Nov 01, 2007 @ 08:47 PM | 2,814 Views
August surgery bought her 2 months, but Amazon's battery finally stopped holding a charge. We have a new laptop on the budget + enough problems with the current systems to keep busy until December.

Foremost, pointing South isn't working. It seems isolated to the heading estimation. Flip on autopilot and heading just locks to a value no matter where she points.

She's had no luck flying cameras because there's too much vibration. The camera must float on the fuselage somehow.

There's still more potential in 1Hz GPS + aerodynamic simulation. The ratio between acceleration & tilt should depend only on throttle. The angle of 0 acceleration should depend only on wind. It should be possible to reverse engineer these factors from the current hardware.

Reading between the lines on autopilot.sourceforge.net, the 4Hz uBlox GPS only does altitude to 0.5m resolution. We've seen that the 1Hz EM406 does 0.1m resolution. We've seen that the EB-85 only does 0.5m. Pretty decided on the Novatel Superstar II at this point....for December.