SMALL - Telemetry SMALL - Radio
Jack Crossfire's blog View Details
Posted by Jack Crossfire | Jul 25, 2015 @ 06:30 PM | 1,249 Views
The Feiyu FY-G3 is what everyone is using to make running videos, nowadays. It's about as cheap as it can be made, when the cost of the motors & the custom compact frame are factored in. If any money was left over after current rent prices, it would be time to make an order, but there is no money & no plan for next year's rent increase.

Its 1 of the new crop of gadgets whose gyros calibrate when it's in motion. Can only imagine it correlates the motion of multiple accelerometers with the gyro motion.


The mane limitation is the god damn Chinese didn't think of putting in a joystick to control it, so there's no way to lock position while retaining the ability to override it. There is a single button which switches between position lock & hand tracking. The button must be held down for 1 second to lock position & tapped to unlock position, not practical for constant trimming over 20 miles.

The biggest limitation is it's not as stable as the home made gimbal in 2013. Ginger Runner confirmed the silky smooth Gootube videos are all software stabilized. Without stabilization, it's as rocky as the home made gimbal after reworking to use UART converters. Still no clues why the rework made it unstable, but might go back to I2C.

It might be worth getting a Feiyu just for its parts, while adding a joystick & using custom firmware. The same parts from rctimer.com are insanely expensive.


It was theorized that a gimbal should stay very close to its starting position, using only rate feedback. Going over the source code, it seemed to be never actually tested using rate feedback only. It always used total angle feedback at a very high gain, but this stopped working after the UART conversion.

The leading theory is no matter how hard you try, there's going to be more latency in converting I2C to UART than reading I2C directly. The UART has buffering delays on both ends, which could be important. It might be worth testing just rate feedback on 1 gyro, with & without UART mode.
Posted by Jack Crossfire | Jul 24, 2015 @ 11:18 PM | 1,921 Views
Amazon.com is transitioning from being known as a store to being known as a web services platform. For the 1st time, there was no new phone, no new kindle, no Amazon Watch, not a single new consumer product to get anyone excited, yet the stock exploded. AWS had always been a growing monster that no-one talked about, but it’s come out of nowhere in the last 6 months to account for all their net revenue.

Of course, stock gains nowadays are mainly driven by share buybacks, algorithms, & the government. Too bad their employees will never have even 1/1000th of Jeff’s 83,921,121 shares, but that’s the new economy. A typical CEO owns 99.9% of the company & the employees own nothing.

When they're bought out by Google for $500 billion, Jeff will invest it in real estate. The employees will end up homeless because they can no longer afford the rent increases caused by the buyout package they earned their CEO.

Every employee must keep their fate in mind, as their CEO frantically tries to sell the company to their nearest competitor. At the same time, employees can't jump ship to join their nearest competitor because that would violate their non compete agreement. Only the CEO can sell the company to their nearest competitor.
Posted by Jack Crossfire | Jul 23, 2015 @ 11:16 PM | 1,112 Views
It requires a Macbook with MacOS already on it. There have been tall tales of pirated MacOS working on a Windows box, but don't believe everything you read on the internet. In this installation, MacOS & Linux have their own bootable partitions, so they don't need a virtual machine to run separately & can get the full machine resources when needed. The trick is getting them running simultaneously, so the development environments of both sides are going at the same time.

Step 1, get Virtualbox to boot the MacOS installation on its existing partition. The magic commands were buried in http://www.virtualbox.org/manual/ch09.html#rawdisk

VBoxManage internalcommands createrawvmdk -filename /path/to/file.vmdk -rawdisk /dev/sda

Creates a virtual disk which is really a symbolic link to the entire real disk.

Attach the virtual disk to the virtual machine in the Virtualbox settings. It worked on the 1st try.


Step 2, fix the screen resolution.

VBoxManage setextradata "mac" VBoxInternal2/EfiGopMode N

Where N can be one of 0,1,2,3,4,5 referring to the 640x480, 800x600, 1024x768, 1280x1024, 1440x900,1920x1200 screen resolution respectively. This only worked with MacOS.

The MacOS booted easily, most of the time.The issues with hidden serial numbers & esoteric bootloaders on the internet were resolved years ago.

Step 3, make the virtual MacOS share the laptop's ethernet so you can access the virtual Mac desktop over VNC & share files over SMB....Continue Reading
Posted by Jack Crossfire | Jul 23, 2015 @ 01:06 AM | 1,270 Views

Comparing edge detected frames from 2 drives with 2 different lighting conditions revealed a slight chance of matching the frames from 2 different drives as a means of lane keeping. With frames matched, it might detect horizontal offset.

An initial test would try synchronizing 2 identical videos of the same drive, purely based on feature matching the exact same frames. Then it would move to 2 different videos. The initial test would match the starting frame with a window of starting frames in another video. It would advance the window of possible matches based on the previous match. It would later use GPS position to aid the frame matching.

Then would come tagging the part of each frame which contains the desired heading, probably manually. Maybe they could be tagged by playing backwards, matching a later frame to a point in an earlier frame to deduce the desired path. Finally would come adding the newest drive to a database of known drives to keep the data recent & to refine the localization.

It's a small step towards everyone's ultimate goal of being able to precisely locate position by searching an exhaustive set of images from every possible location, then refining the database of images with every new localization. It's always been a dream for quad copters, but quite difficult because of the 3D space & the precision required. A path following rover might be a feasible application. The Goog will be the 1st to acquire a startup that can do it.

It's a much more complicated algorithm than ideal & nothing so far has worked, but the initial proof of concept would be bearable. Calif* is predicted to have the wettest rainy season on record. July was already quite wet, by normal standards. There won't be much driving, between commutes & rain.
Posted by Jack Crossfire | Jul 19, 2015 @ 08:14 PM | 912 Views
Continuing the story,

Any robot is an endless series of glitches that must be lived with to make any progress. In the intervening years, there were some attempts to make the 8192CU work more reliably as an access point.

To compile a scratch built kernel for the PI, there was a good document on http://elinux.org/Raspberry_Pi_Kerne..._kernel_source


Then, there was a good rt8192 module compiling document on http://bogeskov.dk/UsbAccessPoint.html

Merely change the Makefile to generate ARM objects using the path to the ARM kernel. Helas, this module was no better than the stock one in 4.0.8+.

There was a nugget about compiling a new hostapd, but the cross compiler didn't have a lot of required header files.

The RTL8192 worked well with the LG Optimus F3 in root mode, but was intermittent at best with the LG Tribute. The only way to start the vision program was using ethernet to ssh in & run

./nohup vision&

That ran it detached from the console.

Wifi didn't stay connected long enough to do that. The phone was definitely part of the problem. Nothing has ever worked as well as manually running iwconfig every second to reconnect in the clear. If only wpa_supplicant could poll like that.

There are signs of a way Android apps can reconnect the wifi connection when it dies.
Posted by Jack Crossfire | Jul 19, 2015 @ 12:22 AM | 1,727 Views


Modified the RC truck cam to make it slightly higher & record uncompressed frames without the wide angle attachment.



Truck cam from February, with wide angle lens & JPEG compression.


...Continue Reading
Posted by Jack Crossfire | Jul 18, 2015 @ 01:35 AM | 1,358 Views








The answer is yes. Developing for iOS is expensive. A pile of I gadgets came from the day job, with plans to support the iWatch. It gets more expensive as the gadget gets smaller, with the iWatch the most expensive, since it requires the phone, which requires the mac.


The 13" macbook from 2012 lasted longer than any other laptop, with no issues in 12 months of daily usage. To be sure, duct taping the keyboard to stop it from glaring sunlight gummed up the touch pad & the screen. Once the gum wore off, it was useable again.


The typical product nowadays is a gadget that communicates with a phone. The ideal situation is MacOS for phone development & Linux for embedded development running simultaneously on a single computer. This was very slow on the 13" mac with 4GB RAM & 500Gig of platters. It couldn't run Android developer studio when running XCode.


The day job provided a 15" mac. It was a modern marvel, 2880x1800 display, 16GB of RAM, 1TB of flash, quad core 2.5Ghz, thin as a razor, costing almost as much as a month of rent. To be sure, more powerful laptops for a lot less money exist, but they never lasted more than 6 months & weighed a ton.


Unfortunately, Linux is a long way from supporting the Macbook Pro 11,5. The 13" continues to run the development environment. The main issues were no wifi & no suspend. It will be obsolete before it's ever supported. Attempts to run Linux on a raw partition in a virtual machine have failed. It might work with Linux on a virtual disk, but this would only be temporary.
Posted by Jack Crossfire | Jul 12, 2015 @ 06:03 PM | 1,769 Views


What the day job lacked in money, it had in bluetooth modules, so it was time to whack together a sliding door lock. Of course, the actual bluetooth board was a trade secret. The servo was a 10 year old piece from the DVD robot. The H bridge was the remanes of the laser projector from 2011, which itself was made from the 72Mhz remote control from 2007.


...Continue Reading
Posted by Jack Crossfire | Jul 05, 2015 @ 05:15 PM | 1,462 Views




This was the ducks guts for getting stereo with cheap electrets.


July 4 fireworks in San Francisco (3 min 44 sec)



...Continue Reading
Posted by Jack Crossfire | Jun 28, 2015 @ 05:13 PM | 2,030 Views
After waking up expecting to see another rocket barely miss a barge or delayed launch, was surprised to see it exploded during liftoff. Of the 3 camera angles, the best one came from NASA TV, then was rereleased with banner ads & annotations.

Falcon 9 explosion from 3 cameras (1 min 13 sec)


The 2nd stage oxygen tank was the scene of many launch delays. Stuck valves, leaky pipes abounded. Not surprising that it finally blew up, probably from a stuck valve.

It's now been 2 consecutive failures in reaching the space station. The last successful mission was April 14. The Soyuz on April 28 failed. The last Cygnus attempt failed on Oct 28, 2014. Only 4 of the last 7 missions made it. The chance of reaching Mars has been better than reaching the space station.

The increasing rate of failures as rocket companies reach ever higher valuations makes you wonder when something has to work to be worth $10 billion & when it just has to be worth $10 billion to prevent a recession.
Posted by Jack Crossfire | Jun 20, 2015 @ 05:45 PM | 1,751 Views




NASA's last space shuttle external tank to complete Endeavour L.A. exhibit

Another win for Calif*. Now only the boosters will be fake, if the exhibit is ever funded. Completion was once envisioned in 2015, then it was 2018, but it was always based on donations & refinancing. If ever delivered, it'll be another spectacle as its transported down the streets of LA. Too bad some water can't be delivered in the tank.


Nearly a year after its inception, the travel fan has been highly successful at defeating the heat. It hasn't chopped any limbs off, hasn't tipped over. Keeping it close & blocking off the area around it has kept kids away from it. Humans are strangely attracted to putting their hands in spinning blades, especially Americans. Limiting the power to 1/4 power has left enough airflow on 12V.

On the downside, it became clear Chinese RC motors aren't capable of running continuously for days. It wasn't long before its RPM fluctuated, then it sometimes stalled & clicked. The bearings failed. Frequent lubrication has kept the bad bearings going.

Also, it's still noisy, despite every effort to silence PWM noise. The commutation noise & shaft resonance are deafening at low RPMs.

The 15 year old squirrel cage blower would make a comeback if not for the fact that it needs to be hosed down with a high pressure nozzle & the outdoor water was shut off.
Posted by Jack Crossfire | Jun 19, 2015 @ 11:33 PM | 2,029 Views
Posted by Jack Crossfire | Jun 18, 2015 @ 01:40 AM | 2,152 Views
Finished it in 32 miles of pounding up & down the trail. The title alone conjured the imagination. Long ago, Carl Segan said one day we would find life on Mars & the martians would be us. The title was obviously a play on that nugget.

It definitely has a few flaws. The mathematical descriptions sometimes get tedious, yet some of his solutions to problems are over simplified & wouldn't really work. It's technical sounding enough to sound like everything is factual. Sounding accurate is what it does better than any other book.

Otherwise, it works. It's easy enough to see how the general direction in the hacking of the various parts could work in real life. It's interesting to see how many different ways the oxygenator & water reclaimer could be basterdized to do virtually any task & to imagine a world where NASA produced stuff that really did work. If the movie is done right, it could invigorate a lot of interest in funding a space program, for a time. A book isn't accessible enough to reach enough people, but a movie might.

There were many attempts to make a movie about traveling to Mars. None got it right, showing how hard the task was. This one might get it right.





In other news, Fitbit finally had their IPO, but more importantly in today's terms, was valued at $4 billion. In creating unlimited free credit, Old Yellen solved the problem of monetizing private data that companies had no right to sell. They would just sell...Continue Reading
Posted by Jack Crossfire | Jun 14, 2015 @ 10:46 PM | 2,128 Views


The decision was finally made to retire the 4 original, low efficiency monocopters & the world's greatest battery charger to the landfill.

...Continue Reading
Posted by Jack Crossfire | Jun 13, 2015 @ 01:57 AM | 1,833 Views
Realtime running form feedback

There was the vaporware http://www.runscribe.com/, which claimed to capture running form for viewing at a later time. Fitbit & all the other uprated pedometers captured step rate for viewing at a later time. The trick is none ever provided realtime feedback. They were all intended to gather data for the company to sell to advertizers, not focused on providing any realtime feedback to the user.

A metronome is very good at providing realtime stimulus for step rate. The same thing hasn't been achieved for kinematics. Step 1 is capturing the body positions. Maybe accelerometers or autonomous drones could do it.

Step 2 is stimulating the athlete to adjust positions. A key stimulus would be moving a leg vertically or leaning at the right angle. Maybe pager motors could be strapped to the limbs. Nowadays, they're called "haptics" & worth many billions of dollars. To a child of the 80's, they will always be pager motors. There are also ways of creating brief pushing & pulling forces, by moving weights. Actuators for training body movements could be too heavy.

A kinematic simulating device would allow you to feel how a famous athlete poses. Famous performances could be recorded & played back through the pose stimulators. Reproducing a dance move with subtle pushing forces from actuators might be extremely slow & tedious, but something repetitive like running might be more practical.
Posted by Jack Crossfire | Jun 10, 2015 @ 12:20 AM | 2,618 Views
Posted by Jack Crossfire | Jun 08, 2015 @ 11:44 PM | 2,378 Views






So after running to the Apple store & waiting 30 minutes for a guy to finish with the 5k monitor, it became quite clear that I didn't have enough money to develop a 4k user interface. Had created a mockup to test out asset sizes, based on downloadable screenshots. It was too small on the 5k monitor. The screenshots were for a completely different monitor that the author had no idea of the resolution of.

The original plan was to make assets double the size of a 2.5k monitor, but the downloadable screenshots showed them using 1.5x the size. In the end, 2x size was probably right. Also, instead of making a fully functional interface, it would be best to just mock up a screen with the desired sizes. It's still completely impractical to design a user interface by loading mockups in an Apple store.

The standard resolution of the next generation of monitors will probably be 5k instead of 4k, just as it was previously 2.5k instead of 2k. The extra pixels allow a complete video frame to fit in a user interface. It makes sense when going from 2.5k to 5k to double the asset sizes. It's also much easier than making them 1.5x. It may be necessary to have double size assets for 5k & 1.5x size assets for 4k.

Unfortunately, I grew up in a time when the best monitors consumers could get were $200. Now, the crappiest 4k monitors are $400. The 5k Apple monitor has a $1900 tag & requires also paying for the embedded computer. Paychecks are getting smaller & rent is going up by amounts that would pay for a new 5k monitor every month. It's 1 of those times you realize the limits of mortality. Mortals can't save money or have everything.
Posted by Jack Crossfire | Jun 07, 2015 @ 01:52 AM | 2,323 Views






After much debate, finally built it. The RC fan shroud was noisy & reduced airflow. A screen which allowed air through might do better than cardboard. The motor is still too noisy, so the debate is on about reverting to the larger motor.
Posted by Jack Crossfire | May 31, 2015 @ 09:00 PM | 3,349 Views
Solar powered rover photovore thing:

These were popular, 10 years ago. They have no purpose besides reusing parts. There are some giant electrolytic caps & solar panels are slightly cheaper than the old days. The mane problem is all wheeled vehicles get stuck. There was a ramp up in photovore videos when they were popular, then nothing more to do with them was found.


Bluetooth sliding door lock:

The mechanics are simple, but the encryption is hard. Keyless entry systems have been around forever, but are actually an interesting problem. You can't just send a fixed packet to unlock them. The receiver & transmitter contain a pseudo random number generator which is synchronized. Every unlock command uses a different code from the random number generator. If an unlock command is missed, the receiver tests a certain number of future codes. The receiver can only miss a certain number of commands before it requires resynchronization.
Posted by Jack Crossfire | May 30, 2015 @ 11:45 PM | 2,761 Views



The 1st video card. It was PCI. None of this AGP, PCI-E nonsense. It's not about the performance as much as the memories. It was upgraded to some ghastly amount of memory for its time. The day it was upgraded, there was a geeky old guy in the same room, slobbering over a dual Pentium II he just received. It was the seed for the idea of a dual CPU. In the meantime between then & the dual CPU, this card went from 16 to 24 bit color. What an improvement.


The 2nd video card was for the dual CPU & had a lot of issues. After this came the era of freebie's which lasted until 2006. The next video card will be the 1st one bought since 1999.

...Continue Reading