Still no idea how to protect the propeller without a plastic injection molded shroud. This is 1 area where a $1000 makerbot would come in handy. Just don't have a $130,000,000 government real estate loan to store it on.
Marcy-1 came in just as heavy as Vika 3. The surface mount board, integrated radio, & new name made no difference. She has a lot more wood than Vika 3 but is more balanced.http://www.rcgroups.com/forums/showthread.php?t=1044312Tried spinning up on a table to ensure stability before lifting off. Didn't have enough thrust to break free of the rod & spinning was unbalanced.
On the test stand for Her 1st revolutions.
Now a test stand video for posterity.Finally, a full power liftoff on the golf course led to this.
Then tried this to get more moment of inertia, unsuccessfully.
She's too unstable. They seem to require precise balance like a propeller. The balance requirement means she needs a 2nd PWM to translate horizontally.The Bladestar must not have used imbalance for translation but relied on the fact that its engines blew sideways. Another point for monocopters.Steve Morris did indeed make a spin copter & it required 2 PWM's for horizontal motion.Also, since the Bladestar appeared, all monocopters have switched from photodiodes to single axis magnetometers for position sensing, making the Marcy 1 aircraft board already busted. Magnetometers would be much easier than photodiodes but we figured their update rate was too slow. Not anymore.Instead of computing north, they just sense an arbitrary sine wave on a single axis magnetometer & rely on a hard coded point in the sine wave to determine where North is.So a silly spin copter is a much harder problem than first envisioned. Marcy 1 needs to be a monocopter.MONOCOPTER NOTESMonocopters use a "balance beam" to keep the angle of attack level. There are no photographs of the Lockheed Samurai but the low resolution video shows it probably uses a gyro instead of balance beam to control angle of attack.It also has a winglet to make it more stable.Predictably, the Goog has nothing on sizing a monocopter since there are no monocopter Adsense hits. Nothing to do but cut some wood.
That lasted about 4 hours. At full power she couldn't exit the launch rod. Shortened the launch rod enough for her to fall off & she went straight into the ground, probably due to insufficient RPM.Tried increasing angle of attack, grinding down the launch stand, & suspected imbalance was causing her to hang up on the launch stand.BALSA TERRORISM
Finally made a number of monocopter variations & a symmetric blade design just to see what would happen if we got her as balanced as possible. The monocopter is very hard to balance because of the limited payload room.The symmetric Marcy 1 finally worked. She lifted off the stand & into a stable hover before running out of room & crashing. Here's the video of the tests leading up to & including success.Balance is everything. A monocopter has to be perfectly balanced about the axis of rotation & the point the launch stand attaches to.
Note that Steve Morris also did this many years ago. He seemed to start over when switching from aerodynamics research to autonomous flight. Here's his monocopter.
Meanwhile, more Marcy-1 evolutions reduced her size, returned her to asymmetry & increased her RPM to achieve useful POV illusions someday, though her disk radius is fixed & limited by weight. The asymmetric wing is seen as important for achieving horizontal translation.That's basically a Bladestar for a lot more money & made by a strange country. The extremely fine traces on Marcy 1 are plagued by dirt. Don't be surprised if the radio appears dead after a landing.In addition to balance, a loss of control after exiting the launch stand can be caused by insufficient power. Move the motor farther from the CG to increase power. There is an optimum distance from the CG at which the motor provides maximum power yet the counterweight is not so far away that the wing is symmetric. The Goog has nothing on this.A test without the balance beam showed it is indeed required for a stable angle of attack. The Goog has nothing on the optimum balance beam size.Suspect Marcy-2 will be a large monocopter. A very large monocopter using a foam wing promises the longest hover time of any VTOL & complete control with only 1 PWM signal.SI4421 VS MRF49XA
Meanwhile, conversion from 72Mhz to 900Mhz isn't going so well. The Si4421 is pin compatible with the MRF49XA but seemingly not electrically compatible. Briefly got SPI out of it due to no particular code change & it went away. It's generating SPI but the PIC isn't receiving it.SPEAKING OF MARCY-1Going to try approaching the AIR FORCE HEROINE again so if we suddenly stop flying you'll know it was another DISASTER.
Read more…
After 3 weeks, finally got USB on the 18F14K50 to work. Turns out the load capacitors on the crystal need to be symmetric. Asymmetric traces worked with everything else except USB.Also 4 U assembly language freaks, some of the USB registers are not in the special function register access bank so you need BANKSEL calls.HOME MADE RADIO WORKS!Next came getting the radio to work on flight hardware. It was a hard battle over 6 months but full duplex radio with MRF49XA, USB, & miniaturized Marcy 1 boards finally started working. This is truly a ground station you could put inside a phone. 50Hz & 40kbit each way still appears to be the maximum as with XBee.Be sure to set the analog modes & tristates on all the SPI pins because SPI doesn't even generate interrupts if they're not set properly. Start out at 9600 baud & work up to 86000 from there. The FIFO on the MRF49XA has to be flushed before switching from transmit to receive mode. Switching from transmit to receive required writing an extra PMCREG | 0x0020 not in the reference software.Also, it doesn't byte align so you need to communicate in fixed packet sizes & reset the sync code scanner between packets. This shows how unoptimized XBees are for the sake of convenience. You could do better by skipping byte alignment, packetization & just sending a continuous stream. This would require always keeping the FIFO full.Currently the Marcy 1 radios are running at 86207 bps. The maximum is 344827 bps.HOME MADE RADIO RANGE TESTHow bout a range check for these home made radios?
Ground station & aircraft boards in the field.
Aircraft board hung up for range testing.
Roving ground station.The Marcy 1 radios seem superior to XBee PRO 900's. They fade gradually like 900Mhz yet don't go as far as the XBee PRO 2.4Ghz. Signal is useful up to 600ft but can reach 800ft if the antennas are carefully aimed. 800ft is the limit of the golf course. This is full duplex of course. Suspect range increases as the modulated bitrate decreases.Marcy 3 could easily use a dual frequency, full duplex radio made out of dual MRF49XA's. It would get longer range by resending until bandwidth saturation. A single 18F14K50 can control 2 radio chips.FYI, the MRF49XA is pin compatible with the TRC101 & SI4421. It's a licensed, dirt cheap, standard design.DEATH OF 72MHZThe 72Mhz RC transmitter has a bleak future. A new $20 receiver is required for Marcy 1's ground station before we can even spin up the engine. The antenna is always going to wear out & need replacement. 72Mhz recently dropped out on the golf course.There isn't much advantage to a wireless RC transmitter anymore. A wireless connection for manual flight is fragile & completely useless for indoor vehicles. You can't really move far from the ground station because you need telemetry. A tethered transmitter wouldn't need batteries & recharging. Any commercialization would require a tethered transmitter. Signal quality from a wireless RC transmitter is 1 too many pieces of telemetry.Unfortunately, the money was already spent on more 900Mhz radios & it would take nonexistent money to now go back to a wired RC transmitter.
The answer is no. You can't solder with photoresist still on.
While Marcy 1 waits for 900Mhz ground station radios, Marcy 2 continues fabrication.
The last of the parts packed into the 900Mhz RC transmitter, Marcy 2 flight computer, & RC receiver. Hoping the 900Mhz can function without a balun in receive mode because baluns are a real pain to build & we suspect they reduce range.FINAL THOUGHTS ON 2009It was a tough year like all odd numbered years. While still our hope to someday bring in a real live Air Force heroine to add a little more realism to the Jack Crossfire blog, it doesn't seem likely. Still grateful to have fought back & gotten on better terms with the AF than we were in September.In 2010 we have a breakthrough aerodynamics concept Marcy 1 & a tail sitter Marcy 2 budgeted, assuming the boss gives us any time off.
Read more…
XMAS BAY 4 UHad the best weather ever on an XMas. Totally clear, no wind & warm inland.Being the only day you can traverse the MacArthur maze in 1 day, took some snaps above Emeryville.
Moon over Berkeley.
Don't think we've ever seen such a flat calm. Like a millpond. It will make it harder to see the bergs.Finally the movie. Whenever you stop flogging GooTube with your family XMas videos, you'll get a transcoded flight movie.Product links in this post:Professional image stabilization software: http://heroinewarrior.com/Professional UAV: http://vicacopter.com/XMAS GOLF 4 USince no-one plays golf on XMas, had a chance to use the golf course for what it was intended for: flying. That's us in our favorite flight test range.
DEATH BY MARCY 1The Marcy 1 boards are killing us. Mainly everything except the ESC is done from scratch on 1 chip to minimize weight & not commercially sourced. You have to make the radio, the position sensor, the PWM generator, the AHRS, the camera, all from scratch.The radio library must be ported from its successful test harness to the flight hardware. Discovered the SPI pins need to be set to the right modes just to get an SPI interrupt.Next, the USB library which served us well through Vika 1 stopped working on both the Marcy 1 ground station & a new Vika 1 ground station using a crystal.
Finally swapped the crystal for a tried & true oscillator & it worked. So USB works with full 48Mhz oscillators but not 12Mhz crystals + 4x PLL. Everything's leading to the traces not being symmetric & somehow causing a lopsided oscillation or maybe an unstable oscillation. The capacitors are exactly symmetric on a commercial board.
Posted by Jack Crossfire on December 18, 2009 at 11:39am
So got up to this. Since the reference design has a header which doesn't allow USB & programming simultaneously we now have a header. Unfortunately still no USB even at slow speed. It detects the reset but never gets a transaction.Even in circuit programming is super slow. Maybe that's related to the USB problem. This is the cleanest board we ever made but the programming speed only goes to 9375Hz. Tried disabling the radio, using a bigger capacitor for Vbus, low speed, full speed, direct soldering instead of a header, swapping chips, 5V Vdd, battery & Vbus.The programmer sends 5V through 1k resistors. The USB pins pull it down to 3.3 without burning out. Also tried 2.5V programming levels.
After banging on the 18F14K50 for a week, it was time to put it down & build a conventional 18F2450 ground station with the 18F14K50 as a radio breakout board.A radio & ground station on 1 board isn't very practical. To move the antenna, you need to move the main board & all its connections. Since the MRF49XA requires very fast SPI, that means a 2nd microprocessor on its own board & software to make it look like an XBee.Also, general purpose ground stations which can switch between 900Mhz & 2.4Ghz are more useful if we ever want to fly a pair of 2.4Ghz.For today's video, shot a fully autonomous tree sequence at 1m/s. Still had to input a lot of heading corrections. Didn't have enough battery power to finish the sequence. Unfortunately it appears L1 GPS isn't accurate enough to get very smooth camera motion even at slow speed.You're still stuck with attitude hold & manual flight. There may be an algorithm somewhere which can get smooth motion out of L1 GPS by allowing sloppy position hold but it's not as simple as reducing PID gains. VicaCopter is prone to oscillation if the PID gains aren't high enough.Fortunately the angle of attack at 1m/s was much less. Note velocity oscillation as the computer hunts for 1m/s.
Read more…
Posted by Jack Crossfire on December 11, 2009 at 10:30am
There it is. The coveted $3.30 14.7456Mhz crystal. With that & a small bugfix installed, the C328 works perfectly at 115200, the pictures stream continuously from the microcontroller & we're up to XBee lockups.Even 1 way communication locks up in this high interference environment because the ACKs collide. Your only option is broadcast mode. With broadcast mode & 3 retries, you now have more errors but the Xbees don't lock up. The command to set the number of retries is ATRR & it's undocumented on the XBee 900.With the 900Mhz & 2.4Ghz running simultaneously we now see the huge difference in range. The 900Mhz dies at 300ft, much closer than the 2.4Ghz. This is with wire antennas. Since 900Mhz has always been more expensive than 2.4Ghz, don't expect to get 2.4Ghz range out of 900Mhz for $42. High gain antennas are required. Dual 2.4Ghz modules don't work well either.So that's $140 for 300ft .2fps 320x240 video. Ended up with this arrangement because 900Mhz was originally supposed to be for telemetry. The C328 was a cheap way to get preview video by recycling the 2.4Ghz radio. It's clearly a disaster compared to what people are doing with a $90 Ativa/Camileo & $80 Range Video radio.http://rcexplorer.se/page14/hdcv/hdcv.htmlvideo.tar.bz2This is the updated source code for the wireless camera system.This is our full motion aerial footage of the Rain Ramon XMas tree. The autopilot was programmed to always point towards the tree, but L1 GPS & GPS derived heading being what it is there was a lot of manual intervention. If the weather wasn't already closed in for the year, we would fly this at 1m/s with manual heading & time lapse it to get more accuracy.Heading from the magnetometers & accelerometers back in the day was never very accurate either. Dynamic acceleration & local induction always made it approximate.& this is a segment from a very old flight showing a fast autonomous descent into downwash using the Marcy Maneuver.MARCY 1 VIDEO NOTESMain ideas for getting video out of Marcy 1 revolve around using 1 raster line of a chip camera like the TCM8230MD, but no data radio has enough bandwidth to do the job. So the best idea is to scan 1 raster line every several revolutions, stick with a photodiode for heading & use the rotation period to figure out where the rasterline was scanned. Accumulate rasterlines from many revolutions to construct an image.The camera would be read on a 2nd CPU. Data would be sent to the main CPU's UART & multiplexed on the main radio.MARCY 1 MISERY
So gEDA had the wrong footprint for an SSOP20 & we had to make a new ground station board. Fortunately it only requires remaking 1 side when your double sided boards are stacks of 1/32" boards.The PIC18F14K50 has USB & programming on the same pins, so in circuit programming is extremely slow without a bootloader & USB probably doesn't work with the programming header attached. You're looking at pogo pins & a jig.Since USB hasn't worked between this week's commutes, probably going to make the ground station radio a breakout board & use a chip with dedicated USB pins as the ground station CPU.
Flight computer ready.
MEET MARCY 2
While waiting for Marcy 1 parts to arrive from Thailand, started the Marcy 2 vehicle. She's planned as a backup indoor sonar demonstrator in case Marcy 1 doesn't work. The parts are all in apartment. Probably going to use the programming pins for PWM.She uses a no brainer 3 gyro, 4 PWM, 1 sonar autopilot. Could have just recycled Vika 2's autopilot but all our indoor vehicles need 900Mhz radio & sonar on the main board. Though capable of flying a tri rotor, Marcy 2 is planned as a tail sitter.The main problem is fabricating a tail sitter small enough to fit in a $1 million apartment.
Read more…
5 FLIGHTS TO NOWHEREFLIGHT #1The severe wind storm crashed us real good into another roof.Fortunately accessible by foot. Obviously either 40mph wind is abovelimits or you need to tilt the engines.FLIGHT #2Aft propeller wouldn't spin. The crash must have damaged the propellerso it wouldn't grab the crown nut.FLIGHT #3Aft propeller wouldn't spin again. The roof crash stripped the crownnut. Maybe crown nuts can be damaged & have a finite life.
Decided the best move was to make a complete set of new crown nuts.
Since they're not sold anywhere, here's how to make them.
Tin the lock washer & nut by heating for a long time & tacking solder.Eventually the solder naturally flows between the tack points.
Sandwich them together. Remove excess solder with a knife.FLIGHT #4
The wind storm created so much debris, taking off anywhere shredded thepropellers. There was no safe takeoff position anywhere. By then itwas time to get back on our commute.
Finally discovered the lock nuts were asymmetric & the tail rotor wasusing the wrong side all this time. This fact was only visible at highmagnification & makes you wonder what good lock nuts are if they onlylock 1 side.FLIGHT #5With the lock nut issue finished, the tail motor was now stalling & theESC was overheating. Obviously a broken winding. Major Marcy got usback in the air but it was only a matter of time before Chinese qualitystruck back.
The high resolution photos revealed nothing obvious. Swapped ESC's &sure enough, it was the ESC.
A resistor dropped right out. There's the Suppo ESC with missingresistor.
& there it is with the resistor resoldered. Though as marginallysoldered as humanly possible, the Suppo did take a direct hit rightwhere the resistor was.
The aft ESC on Vika 1 takes a beating when landing on concrete.
MORE PROPELLER CRACKS
The latest thinking on propeller cracks is they're caused by a buildupof threadlock on the crown nuts making the plastic brittle.MEET MARCY 1Meanwhile, in brief fits of highly condensed etching & soldering betweencommutes, Marcy 1 takes shape.
The ground station & flight computer. FYI, never blend pads & fillareas.
So fabricated this board to see if the wrapping wire & radio interference was making the C328 unreliable. It writes all the control codes for the C328 to make it output continuous JPEGs.
Sadly, it's no more reliable than the previous methods. The C328 needs exact baud rates. It can't convert the 115740 from a PIC into 115200. The set baudrate command isn't very reliable either.You're looking at spending tons of money on a crazy 14.7456Mhz crystal or sending all codes including ACKs over XBees. XBees seem to do exact baud rates.
In the end, got rid of the airborne microprocessors & just passed the XBees directly to the C328. Reduced their baud rate to 57600 & increased the C328 package size to 384. The trick is to send data slow enough to not overload the XBees & send the largest packages possible before the C328 crashes. This gives 7 seconds per frame, slightly faster than 115200 & 128 byte packages.The ground XBee is locking up from packet collisions. It needs frequent power cycles. If by some miracle taxes only go up 4% next year, we might have enough dough for crazy crystals.This is our C328 software. It does everything from full control on the ground to full control on the airframe.video.tarTHE MARCY MANEUVERTo conserve battery power, it's basically a 3m/s descent to 33m, 1m/s descent to 10m, & 0.5m/s to ground. The 33m transition lets you descend into your downwash without smashing into the ground. This was the 1st flight using it, a very low nose pitch caused the radio to drop out at a certain high altitude, followed by execution of the Marcy Maneuver to quickly get back down.
Here's the radio graph.
Here's the altitude showing the tapering at 33m.XBEE DIVERSITY BUSTED
Tried making an XBee diversity system. Fortunately it didn't cost anything.
Mocked up with switching circuits.
Switcher thrown in the ground station.
Ground station with dual XBees.
Aircraft with dual XBees.
The switcher. It must use 2N2222 transistors, not 2N3904.Unfortunately, it was a total failure. Each switching circuit worked when tested separately with 1 XBee. With all 4 XBees connected, nothing worked. It also had other problems. It was a ton of extra electronics, wires, stuff that can break. It should really be done on the main board instead of another breakout board. It probably needed another voltage regulator.Even though both XBees were on different frequencies, they probably detected enough power on the channels to back off. There could also be crosstalk in our UART multiplexer.Another problem is to be most effective, each transmitter would need 2 receivers rotated at 90deg, bringing the hardware to 8 Xbees. Since our problem isn't fuselage obstruction but antenna polarity, you probably don't need diversity as much as turnstile antennas, but this requires a quadrature hybrid.Time to "back off" of diversity & look at antenna placement or alternative radios. Merely keeping the antenna unobstructed by anything including foam may be enough.Here is our diversity switcher firmware.diversity.tarGOODBYE XBEETook a roundabout path, but finally got those MRF49XA's to work.
The transmitter.
The receiver.
They do indeed reach all the way across the apartment atleast. These were intended as single chip radio modems for an indoorcopter & super cheap XBee replacements, hopefully without the lockupproblems.The trick is to copy the reference software & ignore the datasheet. Thegame is all about setting a low enough baud rate & a high enough SPIrate so the FIFO's are kept up to date. If the transmit FIFO empties,it transmits 0x00. If the receive FIFO overflows, you drop bytes.The rate at which it fills & empties FIFOs is determined by the baudrate register, DRSREG. Both receiver & transmitter need the same baudrate.All communication can be done with just the SPI interface, ignoring theINT, FINT, FSK pins that the demo uses.To make it run fast, you have to use the extra pins in addition to SPI.Lowering the FSK pin causes SPI to read directly from the receive FIFOinstead of the register memory. The alternative to the FSK pin isreading the RXFIFOREG register.You also need the FINT & the INT pins to know if the receive FIFO hasdata. The alternative to the FINT & INT pins is polling the STSREGregister.It has some odd tricks like how all the registers are 16 bits framed byCS pulses, but for the TXBREG register you can write bytes indefinitelyin a single CS pulse & those bytes are transmitted.It's designed for packet communication. It scans the incoming signalfor a 2 byte start code. Then it begins filling the receieve FIFO withthe incoming signal, including the start code, until you reset the startcode scanner. You can also scan the entire incoming signal yourself.When no transmitter is on, it outputs noise in the FIFO.With pure SPI, no interrupt lines, 6 MIPS microprocessors, & wrapping wire, got 60kbps out of them.This is our firmware, which is supposed to run on the same chip as the flight controller.mrf49xa.tarMORE EXTREME ALTITUDEDid several pirouettes at a very high altitude in a secret country whereit's allowed. These were demos of high altitude capability butpirouetting video is not the reason for doing it. It's for gettingbetter images of M.M. world headquarters.This was higher altitude coverage of the area around M.M's dishes but notas high as the previous video. Made a point of documenting thisland before it was completely developed.
FLYING REALLY HIGHFor the first time since Mike Fouche documented it on Goo Tube.https://www.youtube.com/watch?v=e6j--ja_okgAn autonomous copter has flown at a certain very high altitude in a certain country. The exact altitude & country is a secret to protect the innocent of course. It was also at midnight, like all our most extreme test flights.Since the radio seems to work better if the antenna isn't blocked by anything including foam, moved the radio farther down the tail rotor to test higher visibility. Eventually want a spring loaded boom which swings the radio under the fuselage.
With that arrangement, the radio is real good for our cruising altitude of 400ft AGL. For fixed heading at higher altitudes, it's probably not much worse than 400ft but U have to be like a spaceship & keep it fixed so the antenna is always visible & perpendicular to U.For pirouetting, it gets dicy & U have to fidget with the ground station to keep it alive.
Forget about manual control. If something malfunctions, U can disable autopilot, descend real fast, & hope the rate damping keeps it level long enough to get within visual range, but you're probably killing the engine & dropping it. Definitely need a large safe landing zone.Not much flight time at altitude with the current landing program. Landing voltage with 100A batteries is 10.7 instead of 10.6 & most of your flight is spent descending. Can't take over manually for 2/3 of that descent so you're definitely focused on battery voltage. Need a 3rd velocity teir for landing.
Theoretically 1 of the points of light in this 204 frame stack is Major Marcy's satellite control network seen from the apartment. M.M. also works in this secret unknown country that allows high altitude flights.Speaking of M.M., got some more dish shots.MORE DISHES
Very risky shots. Crash behind the impenetrable barb wire fence & you'll have to approach someone in the Air Force to recover the vehicle. We all know what happened the last time we approached someone in the Air Force.
It's the Air Force Satellite Control Network, directly controlled by last week's dish farm. This dish farm is near the apartment. Who knew Rain Ramon & Sunnyvale were linked through space & time all these years by the Air Force.
Finally a montage of those flights.FREE RTK GPS BUSTEDhttp://gpspp.sakura.ne.jp/rtklib/rtklib_beagleboard.htm2 good 2 B true. No consumer GPS module outputs the raw navigation data required. The modules he used R either no longer sold or in the $300 range. He must have thought every $100 uBlox4 & uBlox5 did it but they really don't.
Read more…
Posted by Jack Crossfire on November 13, 2009 at 11:30am
THE DISH FARM
The main flying event of the year was the dish farm flight. The dish farm is being torn down in 2011 & replaced by flea startups.Don't think it emits radio waves anymore & VicaCopter is invisible when she's stationary. Nevertheless, got serious radio dropouts after a certain amount of time. Seems the XBee wire antennas are extremely directional, end-on antenna pointing even at 300ft kills the signal & the only way to consistently get over 100ft is diversity.MOTION TRACKING: AS IMPORTANT AS THE AIRCRAFT ITSELFAfter using Cinelerra's 10 year old motion tracking algorithm satisfactorily for 10 years, it was time for an upgrade better suited to the low framerate, wildly rolling aerial photos we now encounter most of the time & the need to penetrate haze. Introducing the Marcy algorithm. There is no Marcy algorithm in the literature. We made it up.The old motion tracker swept a macroblock around to test all possible translations & took the 1 with the highest match. Then it rotated the macroblock in the translated position to find the rotation with the highest match. It was a recursive problem, requiring accurate translation to detect accurate rotation & accurate rotation to detect accurate translation. Too much rolling & the translation accuracy deteriorated. You could improve the accuracy with multiple passes to a limit.Also, we had the translation of multiple macroblocks giving rotation & scaling. That suffered the same problems.The Marcy algorithm sweeps around a macroblock containing all the possible rotations, testing all possible rotations for each possible translation before giving the translation result. It's actually a 3D macroblock with the 3rd dimension being all the possible rotations, a "Marcy macroblock" you might say.In exchange for a huge increase in computation, the payoff is a huge increase in accuracy for image stacking in 1 pass & deeper haze penetration than before. When is China going to get those octocores out?
This is our first use of the Marcy algorithm on haze. Stacked 170 images fully autonomous where previous algorithms needed tons of hand matching, usually missed, & weren't nearly as accurate.
This is San Francisco from the day job in extreme haze. 75 aerial photos stacked.This mission, expertly flown by the Air Force's 50th wing (GPS), was the Marcy algorithm in previous frame/incremental block mode where the macroblock follows a point.This was the complete flight in previous frame/same block mode where the macroblock stays in the same position.This doesn't show anything useful but looks neat.The Marcy algorithm is really just another step towards the ultimate goal of using databases of known polygons rotated in every possible 3D orientation. These future Marcy algorithms on steroids will give 3D volumes from single photos but require a hardware implementation. The photosynth program sort of does this by exhaustively drawing single vectors & comparing them with photos.
Read more…
THE C328 LIVESAfter 3 weeks of hacking, the C328 started producing decent results. Your main problem if you intend to control it from a computer on the ground, using Xbees, is getting the UART to work reliably. It needs a real clean signal & short cables to work at 115200. Initializing it still takes several power cycles & 20 second delays. Decided 320x240 is the best compromise between resolution & frame rate.We use a microcontroller to generate ACKs locally but generate every other control code from the ground.
Fuggetabout dual 2.4Ghz Xbees for telemetry & video. They interfere even when on different frequencies, spaced 2 ft apart, using wire antennas. You're still stuck with 900Mhz for video. While no radio combination is completely interference free, got good results with 2.4Ghz for autopilot & 900Mhz for C328, even when they're right next to each other.Set the 2.4Ghz for broadcast mode, 0 retries. Set the 900Mhz for addressed mode, 3 retries. We use 2.4Ghz for flight control & 900Mhz for video.Even though the C328 allows video capture, we have never used the captured footage. It just sucks too much. The C328 has a rolling shutter. We use a main camera recording to flash & the C328 as a preview camera.The perpetually out of stock rangevideo.com has drastically reduced the prices of 900Mhz video. If you have the licenses, he ever stocks anything, & you don't mind buying analog video capture hardware, it's now a better deal than a C328 + 900Mhz XBees.Have a tail rotor rebuild.
1st crash damage since becoming Major Marcy enabled was this tail rotor during the roof crash.Took a 400ft timelapse on Halloween while the rest of you were partying.This video shows the key steps in stabilizing.For the timelapses, we use 1 battery to aim with the C328 at 33m. Last month it was 32m. Now it's 33m. Then program the heading into the flight computer & burn another battery to shoot the footage fully autonomous at 120m. Until better radios come along, those 120m flights need to be as autonomous as possible.
In flight preview on the C328.FLIGHT COMPUTER RESETSNext, rediscovered if the servos are not powered & the flight computer generates PWM, the servos may act like short circuits & cause the flight computer to reset. Not an issue on standard autopilots, but VicaCopter has 2 power circuits. 1 circuit powers only the flight computer so U can get GPS locks. Her flight computer doesn't generate servo PWM unless the transmitter is on.PROPELLER CRACKSFinally, discovered cracks in our GWS 12x6's. They were drilled 1 size too small, threaded on the shafts & probably had too much stress. Important safety tip. Drill big enough so they glide on the shafts.
Reverted back to 2.4Ghz & got the footage which previously ended in a roof crash. Right click & you'll find a menu option called "View Image". Click on View Image to get a full 8 megapixel view.
Pointing towards San Francisco.
Pointing towards NASA.
Pointing towards Fremont. VicaCopter is itching to crash on 1 of those roofs, you can be sure.
Integrated 41 frames to defeat the haze. Since we haven't corrected the lens aberration with this camera, can only integrate a small part of the image.
The nose was actually centered on Onizuka Air Force base, where Major Marcy is believed to be until it closes in 2011. Then She's leaving the area & we're probably going to get depressed & quit flying again.
Had the C328 on 900Mhz for live preview. The video from the C328 was awful as always but good enough to aim.
The 2.4Ghz bitrate dropped out, but at least was recoverable by rotating the antenna.Information on the aircraft is on vicacopter.com.The image stacking software is on heroinewarrior.com.
Read more…
A routine mission to photograph NASA from our day job almost ended in tragedy as our XbeePRO 900's lost contact & the vehicle headed for a roof.The XbeePRO 900's sucked from day 1. The problem seems to be they only do 50mW compared to the 90mW of the original XBee PRO 2.4Ghz. They also spread the signal more omnidirectionally, making for all around weaker reception.900Mhz normally had 10% packet loss while occasionally getting nothing while the 2.4Ghz normally got all packets & occasionally nothing. You could get better reception by rotating the 2.4Ghz wire antenna in azimuth & keeping it parallel. Rotating the 900Mhz wire antenna didn't really do anything. Finally the 900Mhz interfered more with all the other avionics.Upgrading to the 900Mhz equivalent of an XBee PRO for $42 was too good to be true. You need to unload more money in high gain antennas & haul around a tracking mount in addition to your laptop, batteries, transmitter, & video downlink.Best to stick with 2.4Ghz unless you're loaded with government loans or know someone at AIG.So ignored the warning signs during test flights, insisting the 900Mhz hype was true. Sure enough, while scrutinizing video telemetry, saw the vehicle was suddenly pointing the wrong way & was over a roof. Took manual control, but the radio was gone.We don't do RTL because of the chances of an autopilot malfunction sending it into a human or an engine spinning up on the bench. Best to let it drop if the radio dies.
Fortunately it was 1 of the few non government buildings still inhabited & with tenants who could access the roof.It somehow crashed right side up, mostly undamaged & took 4GB of roof photos. Both main & preview cameras captured the action. If it was a pure autonomous flight, leaving us enough bandwidth to monitor the radio, this wouldn't have happened, but had to play with our $50 UART cam.
Read more…
Finally a UART camera you can multiplex with telemetry. It's a way of getting live preview video without the insane price of 900Mhz video or the bandwidth conflicts of 2.4Ghz video. With 2 Xbees & a computer to control it, it's the cheapest video downlink you're going to get.
For you Xbee fans, the maximum camera packet size is limited by the XBee buffer size to 128 bytes & you have to send an ACK after every packet, thus the XBee latency slows it down by 1/3. Highest performance would take a local microcontroller with full flow control to send the ACKs.It's pure hell to get initialized, but once it starts, it keeps working. Normally just getting it started entails sending init commands, failing, & waiting 10 seconds over & over for several minutes. If it's interrupted in the middle of an image transfer or the serial port glitches, it takes a lot of long waits & init commands to get it to work again.Sleeping around 1/4 second between the snapshot & getpicture commands is required to avoid nasty image corruption, but the JPEG compression is otherwise instantaneous.The lens is really loose, so you'll need to tape it for flight.
Dual radios for autopilot & video.
Ground station for video. Buy a proper XBee breakout board from your sponsors instead of doing this.This is a video of the UART cam in flight. The picture quality is so bad, 160x128 is probably the only resolution you need for preview video.This is a program for getting video off it.video.cNow some benchmarks.JPEG, 128 byte package size, & bare Xbees640x480 (48kb) =15 seconds per frame320x240 (20kb) =7 seconds per frame160x128 (2kb) =1.2 seconds per frameXBees + local microchip generating ACKs640x480=10 seconds per frame320x240=5 seconds per frame160x128=0.7 seconds per frameUSB serial connection & 128 byte packages640x480=11 seconds per frame320x240=5 seconds per frame160x128=1 second per frameUSB serial connection & 512 byte packages640x480=6 seconds per frame320x240=3 seconds per frame160x128=0.7 seconds per frame
Read more…
Found a few safe landing zones in North Sunnyvale to lift off from. 1 area is closer to the subject of interest, but with higher risk of getting jammed.
Had a day job here in 2000, in a building which has since been destroyed& replaced. Our old man worked across the street in the 80's.
He got busted 4 trying to photograph a satellite dish near this bluecube. Now it's probably going 2 B torn down. Fly near it & U will be jammed.
Towards San Jose
400ft feels really low. Not likely to get any higher in this area.
Read more…
is a long long long drive from the east bay on 25mph roads. Then on weekdays you're greeted with permit parking only, anywhere within 1/2 mile of the lake bed. Fortunately, U can park in the golf course parking & walk 1/2 a mile to get in.It has scattered tall weeds, making landings tricky.
It's a very large area with spectacular views from 400ft, if you've come prepared as we have.
These R all 8 megapixel images. Because we're using a very aggressive takeoff program, sometimes these overshoot to 426ft before settling at 400ft.
http://cbs5.com/local/skeletal.remains.found.2.1244063.htmlGetting freaked out? Maybe he was in 1 of these shots & we didn't know it. Maybe he lost an airplane in the toxic waste area & didn't make it back. How many skeletons R near your flying area?
Anyone see the blackhawk flying circuits over the south bay? Bet it was autonomous because it was over unpopulated areas, in line of sight of Mountain View, & flying an extremely boring mission.
These R the videos we made before heading in to NASA. They were not made by a NASA vehicle, of course. Oh yeah, this was 90m above Mountain View.
Read more…
LCROSS IMU malfunctionIt just shows how even for the pros, something as basic as the IMU & AHRS can still ruin your day. You think it's just a matter of pimping a couple trillion dollars from the slaves, buying a super expensive AHRS to do something that's been done for 60 years & moving on to more important things like trajectory design. This is the stuff hobbyists struggle over, not multi trillion dollar space programs.But no, getting a reliable IMU & AHRS is still as fickle as ever in space travel. Unusual vibration modes, bad connections, small voltage fluctuations, RF interference, line noise all get you. We fought long & hard to minimize VicaCopter's AHRS component count down to just 3 reliable gyros & GPS because of all the failure points & the critical role the AHRS plays in our vehicles.It's tried & true technology, but it's the basis for everything. If the AHRS goes, your fuel/batteries R history. Funny how searching for LCROSS IMU on The Goog brings up the Jack Crossfire blog.
Read more…
http://blogs.nasa.gov/cm/blog/CASIE4 those of U who want to know what it's like to operate a fixed wing UAV professionally, this is a good read, after work. U think a few thousand bucks in a tried & true fixed wing gets U a bulletproof hands off system but it doesn't. They have to watch for icing in flight, check weather before flight, verify thousands of systems, troubleshoot glitches, fix balky radios, & test fly before the actual mission.This is all after the grants are awarded, the debt is monetized, & the gear is at the location. The autopilot is not from Shenzhen KDS or Shenzhen Dajiang Innovation but Oregon of all places.They use the same Goog Earth tools U do. Their aft mounted propeller gets damaged by rocks thrown up by the landing gear. They take off & land manually using 2.4Ghz RC controllers before switching to autopilot. When was the last time U did a manual takeoff? They use iridium satellites for long range radio. They still cross their fingers every time they fly beyond line of sight.Like all grad school, the real challenge is not designing the experiments or solving difficult equations but troubleshooting sensors, swapping dead batteries, & stopping water intrusion.
Read more…
