New Products Flash Sale
Jack Crossfire's blog
Posted by Jack Crossfire | Nov 18, 2008 @ 03:30 AM | 4,879 Views
Now that we're not flopping anymore, can see the quad is still capable of nasty vibration. Propeller balancers will be in attendance.

Decided to forget about manual cyclic & begin implementing a full time orientation hold & switchable position hold on top of that. That is a buster to get all 3 modes to coexist simultaneously for quad rotor & single rotor craft.

With enough software functioning between commutes, test flight #2 was still disasterous. Cyclic rate damping is good enough for the computer to hold the attitude, but still not for a human. There is no yaw authority. Yaw damping is ineffective & it takes very large deflections for the computer to try to hold a heading, which it never quite does before 1 of the 2 motor pairs stops.
Posted by Jack Crossfire | Nov 16, 2008 @ 05:18 AM | 4,938 Views
Looks like all quad rotors have to overclock their PWM way above the usual 50Hz. Mikrocopter uses I2C to get still higher update rates. The answer is yes. Suppo/Super Simple ESC's can do 100Hz, but it requires shrinking the resolution. The standard resolution only goes to 60Hz.

Nonlinear feedback curves, lighter propellers, higher gyro bandwidth, & higher frequency PWM has all contributed a bit to stability. It's never going to be as stable as an articulated rotor.

In today's video, U can see the rate is damped enough for a human to hold the angle, but the stick response is prone to violent rotation.

Quad rotor rate damping 2 (1 min 35 sec)

Forget about having enough money for any new parts in November. The Suppos R what U get.
Posted by Jack Crossfire | Nov 14, 2008 @ 01:50 AM | 5,187 Views
Well, Microdrones raised the bar with an 8.5lb quad rotor capable of 50 minute flights on battery power. Dual redundant IMU's & flight computers. Must figure out how Germans got so loaded compared to US.

Anyways, back to bailing out SUV manufacturers.

Decided blogs based on user content R really boring now because no-one's buying anything but T bills, so here R some rate damping efforts. Unfortunately, Goo Tube audio synchronization doesn't convey the relation between movement & throttle changes & sure enough, Goog stock is only $300.

Quad rotor rate damping (2 min 35 sec)

See. No rate damping at all. Disappointed to find the "New 1000mm Quad Copter Design" was using Turnigy ESC's at only double the price, not the 10x price U would expect for the ESC to be the problem. The next step is really full time attitude hold. Everyone uses full time attitude hold. Disable lowpass filtering completely & U start to get some rate damping but it's very erratic.

Also, these props R 72% heavier than successful quad rotor props. Haven't sold the T-Rex yet.
Posted by Jack Crossfire | Nov 13, 2008 @ 03:29 AM | 4,946 Views
That didn't go so well. Throttle control isn't fast enough for rate damping to work at all. It's a complete, unflyable, quad disaster. The Suppo ESC's either don't have the response time or don't have enough throttle steps. That's what U get for being sucked into super low prices.

There's much less vibration in the quadrotor than the unirotor. With virtually no noise in the IMU, U don't have the drift anomalies that plagued us & U can probably use the accelerometer for some velocity estimation.

Not surprising that with all these advantages, Mikrocopter has solved most of the problems of autonomous VTOL in the last year & you'd now be hard pressed to bother with all we did when U can just buy a Mikrocopter. They're hauling in some serious dough with those parts.

Mikrocopter does it all on an 8 bit Atmel. It needs no Kalman filters, quaternions, ground based computing, crazy communication schemes, neural networks, or anything. Just straight gyro integration on all 3 axes, using integer math. Unfortunately the source code is in German & it depends on some crazy fixed point optimizations, so your hacking mileage may vary.

If the dynamic forces are low enough, a very high accelerometer blend & slow yaw rate can eliminate the need for a quaternion attitude estimate. High wind still might give U problems. Suspect quaternions & Kalmans will come to Mikrocopter soon enough, which is the biggest reason to still bother with VicaCopter:...Continue Reading
Posted by Jack Crossfire | Nov 12, 2008 @ 06:02 PM | 4,203 Views
When spare parts for our Corona dried up, it seemed Lite Machines was gone. They shifted completely to an extremely painful suppository shaped VTOL device that seemed doomed.

The biggest problem was the requirement for hand launches & landings. For years, no-one was willing to reach into the whirling mass of blades to grab it, but thanks to some recent bank layoffs, that's not a problem anymore & now the mighty Lite Machine VTOL lives.

Lite Machines' Voyeur VTOL UAV - the real thing (1 min 2 sec)

Posted by Jack Crossfire | Nov 12, 2008 @ 02:47 AM | 4,259 Views
PIC debugging complete. There was a bug where DAMPING_SIGN variables interfered with analog readings, but just moved the DAMPING_SIGN variables somewhere else & it started working. Maybe it'll explode over some yellowjackets.

IMU calibration complete. Throttle range calibrated. Rate damping & control directions tested. Engine cutoff & failure testing complete. Threadlocking applied. Ny-ties tied. LED's lighting. Went back to engine cutoff switch + throttle stick since there isn't enough off margin when the transmitter is on & the rate damping is active. Now those propellers R really off when they're off. Running out of things to do without flying.
Posted by Jack Crossfire | Nov 10, 2008 @ 01:49 AM | 4,625 Views
The answer is yes. There is now a yellowjacket hive behind the dumpy apartment complex. If U go there to photograph or fly, you'll be stung over 10 times by a swarm, mainly in the head & have excruciating pain for the next 6 hours, followed by bearable pain. You'll need to run back to the nearest water supply as fast as possible & spray yourself in freezing weather to try to get rid of the monsters.

Still had enough unaffected skin after the attack & extreme pain to finish the quad assembly. Engine testing is already disasterous. Propellers spin up randomly from gyro noise. Prop savers break. Nuts & washers R falling from the fuselage.

Quad rotors R real dangerous. Unirotors don't have any starting torque. Quad rotors can go from 0 to full RPM instantly. Battery connector is too close to the rotors....Continue Reading
Posted by Jack Crossfire | Nov 08, 2008 @ 06:24 AM | 4,468 Views
Getting the quad rotor powered up is going to take another week. Going to 4 engines makes things very complicated & unreasonable. It's a matter of commuting back & forth every 2 hours, learning as much as possible between commutes, getting a missing part or tool down in the valley. We have 1 defective prop saver.

To test the rate damping, need to unplug all but 1 of the motors & wiggle it bare handed with the propeller spinning. IMU mounting isn't going as well as planned either. The total cost is heading more into the $250's than the $132 envisioned....Continue Reading
Posted by Jack Crossfire | Nov 06, 2008 @ 03:23 AM | 4,491 Views
the quad rotor.

Motor mount & mockup #2 was sturdier but not perfect. Final assembly will be foam tape intensive. The plywood fuselage is soft indeed. Time to fabricate this: 4 hours & it's back in the car & back to Silicon Valley.
Posted by Jack Crossfire | Nov 05, 2008 @ 03:18 AM | 2,992 Views
Enough parts arrived from China to mock up with actual motors & materials. The motors were half the expected size & really really flimsy. Looks like our revolutionary theory of metal wires clamping motors to CF rods is busted. It works for a stationary object like the contact lens agitator magnet, but not an aircraft. Looks like it's going 2 B big, heavy wooden vices like everyone else.

Only got CF rods because they were the cheapest & strongest. Need to crash what we have before going to proper square tubing.

Speaking of revolutions, how bout that human nature? U don't need 2 B a psychology major to predict how humans will respond to mass media. Your world is going to be a lot more like...Calif*. If U don't know what Calif* is like, U haven't been reading the Jack Crossfire blog. 450 sq ft of living space is going 2 B a victory 4 U indeed. Better get rid of those savings accounts....Continue Reading
Posted by Jack Crossfire | Nov 04, 2008 @ 04:11 AM | 2,724 Views
Quad rotors R a return to the past.

Our first campaign to build a UAV anything was a quad rotor in Fall of 1988. The quad rotor would be solar powered & recharge a bank of NiCd batteries. It would use discrete electronics & brushed motors. It would be made of a sheet of plywood. It would fly under FPV control using some kind of wireless camera that we would design, & signals sent over 2 way short wave radio. If the NiCD's didn't have enough capacity, it would land & recharge several times during a mission.

Its mission was to fly 12 miles over the east bay hills & bring U telepresence from your true love. You see, we didn't have cars in those days. Also didn't know about microcontrollers, rate damping, counter rotating props, camera systems, wireless systems, brushless motors, solar panel efficiency, battery efficiency, amputation by propeller, & economics. However, it was legal in 1988.

The technology & economics wouldn't exist for another 20 years. In 1988, it might have been barely possible if we used a motor glider or ground vehicle, but our quad rotor compassion would not be denied.

The mission could probably just barely be done today. It would take a very very long time. The quad rotor would fly 15 minutes at a time, recharge on solar power for 7 days, & fly another 15 minutes. It would have to be very autonomous. It would use the cell phone network for video & control, but be entirely GPS guided.

Nowadays we do that exact 12 mile mission every day in a car for our day jobs & the true love is long gone. Ironically, after 20 years of gas & single rotors being economic reality, the electric quad rotor we originally envisioned has emerged as the most practical vehicle.
Posted by Jack Crossfire | Nov 03, 2008 @ 02:02 AM | 2,615 Views
Should note it was only 13 months ago when our vehicles started hovering autonomously. Oct 10, 2007 was the first time anything stayed in the same position on its own. 1Hz GPS wasn't fast enough to do the job, yet neural networks created just enough inferrence to keep it in the air. It was rough going. Forget about any wind. The first year was all about increasing the maximum wind speeds, waypoint tracking, & bringing up airframe #2.

It was only 9/22/06 when we got anything to fly at all: the Picco Z. It was only Oct 24, 2006 when the Corona started flying at all....Continue Reading
Posted by Jack Crossfire | Nov 02, 2008 @ 03:07 AM | 2,880 Views
It's a race against the Chinese air mail, which is probably on a bus. Software migrated from single rotor to quad rotor. 5V leads removed from servo headers. Quad electronics tray being designed.
Posted by Jack Crossfire | Nov 01, 2008 @ 02:14 AM | 2,753 Views
The remains of the Rex might be 4 sale or might not be. It's completely airworthy except for the electronics. If it is ever sold, the electronics will be removed. The tail struts, flybar, & paddles were barely used. The flybar was crashed once. The foam & carbon landing gear assembly is flightworthy & probably stronger than the Taiwan version.

A new kit with ESC, motor, 2.5Ah battery, & CF blades is $449.99.
Some parts missing from VicaCopter's remains:

2.1Ah Battery: -$55.99
Motor + ESC: -$62.50
CF blades: -$36.99
Landing gear struts: -$6

Most of the stuff is used yet there R a few new & spare parts, so the remains of the Rex would probably be $150 today. Don't feel our main blade collection is airworthy enough to sell.
Posted by Jack Crossfire | Oct 31, 2008 @ 04:10 PM | 3,120 Views
Paypal came through & managed to suck our last paycheck into China, so forget about the VicaGlider budget & get ready 4 VicaCopter III.

Now an embedded update.

Phone fans don't like talking about the CPUs in their corporate branded trophies because they're so damn slow. That's why it was really hard to find this out.

iPhone 3G: 412Mhz ARM CPU No FPU.
gPhone: 528Mhz ARM CPU No FPU.

The 600Mhz Gumstix boards we used R being phased out & replaced by new green themed boards running at, wait for it, 600Mhz! "NEON" FPU.

While advertised as dual core, the gPhone CPU is really a 1.5 core & the half core is for communication.

gPhone is written in Java. iPhone is written in C. The gPhone GUI is less responsive. Corporations don't pass higher clockspeeds to users. They use the clockspeeds to write cheaper software.

So unless you're into programming DSP's & FPGA's, the embedded floating point hasn't gone anywhere.
Posted by Jack Crossfire | Oct 30, 2008 @ 04:03 PM | 3,348 Views
Well, $12 motors R a bit too good 2 B true. U need to pay $30 shipping from China to have them in your lifetime. That is, if U can get your credit card to fund Paypal. Anything beats calling the Bangalore call center & waiting on hold for an hour to unblock Paypal. So unless you're crazy, the motors R $50.

Then of course, there's VicaGlider. Since the quadrotor requires parts that only exist in China, we expect major down times. We've got major mortgage bailout taxes coming up. Because of the difficulty in getting money to China, the overlapping quadrotor & VicaGlider purchases could take 2 months to finish, making the overlapping purchases more like 1 purchase each month.
Posted by Jack Crossfire | Oct 28, 2008 @ 11:30 PM | 3,850 Views
Well, the revelation about $12 motors changes everything.

It's time to put the T-Rex down. The starboard servo is repeatedly failing 4 a reason & that is either a damaged BEC or extreme sensitivity of the HS-65 servos to stalling.

The hope was to get a full life out of all 4 servos until burn out, but a replacement servo & BEC is 1/2 the cost of a complete quad rotor.

All crashes with the T-Rex were servo wear, radio failure, & tail belt failure. Quad rotors should never wear out since they have no brush motors. Assume U can control the radio failures & U have no reason to expect the quad to crash in normal operations.

Brushless motors + ESC's: $22 * 4
2 pusher props: $8
2 puller props: $2
Carbon fiber rods: $20
Total: $120

The economics don't support articulated rotors anymore. Our recently federally funded bank's credit card seems to have reset for November.

Now some secrets of lwneuralnet. It might be hackable to support backpropagation through time.


// solve network forward in time

// compute error gradient for network backward in time

// Need to get the derivatives from the inputs of one net_train_batch to
// the outputs of another net_train_batch.

// Set derivatives

// Adjust weights based on average error gradient for all time
Posted by Jack Crossfire | Oct 27, 2008 @ 10:12 PM | 6,807 Views
The latest starboard cyclic servo failed after 5 minutes. It was the aft cyclic servo which we gimbal locked & overheated earlier. There's also a theory that the CC-BEC was damaged in a crash & is burning out the servos 1 by 1.

Well, the T-Rex 450 has been so problematic, plans R already being drawn up for yet another airframe in 2009. That would be 3 airframes consumed in operating 1 UAV. Still more efficient than mortgage stimulus packages, as hard as it is 4 a certain country to believe.

The next airframe would be a quadroter. That's right. Except for the motors, quadroters can be built out of cheaper parts. Quadroters R perceived to be safer. Quadroters have more space for electronics. The main concerns R flight time & damage to expensive motors instead of cheap landing gear. There would be head on crashes into motors, bent casings, bent shafts, broken props, & bank failures.

With all these crashes, now would be a good time 4 U 2 view the Nevada test site....Continue Reading
Posted by Jack Crossfire | Oct 26, 2008 @ 08:04 PM | 3,451 Views
Well, it's cheaper than Taiwan. It shouldn't roll over as easily as Taiwan. Not sure how increased vibration is going to affect the connections. No plans for a camera.
Posted by Jack Crossfire | Oct 26, 2008 @ 01:24 AM | 3,561 Views
Step #1: Get rid of dependance on Taiwan slaves for landing gear.

Step #2: Reduce size of electronics tray to reduce turbulance.

Now some other ideas.

Neural network preprocessor: instead of using for loops & dynamic arrays like the current libraries, generate unrolled SIMD assembly language routines for networks defined at compile time.

Backpropagation through time: not as sexy as evolution & hard to

There is source code for backpropagation through time. It's all from 1993 & can't compile on modern systems.