So the PX4flow issue ended up being the UART overflowing. The floating point conversion must have slowed it way down. IMU drift still seems within limits. There are a few more UART routines to optimize for the higher latency, a few more glitch crashes.
After all that floating point rewriting, DCM conversion, & IMU hacking, the result was still pretty bad. Optical flow is still a sloppy method, especially outside the 1m fixed focus of the PX4flow.
It would scream if it had an autofocus based on the altitude data, but there's still no way to make a long macro lens with autofocus down to a price.
Getting the video feed on Qgroundcontrol requires disconnecting the PX4flow from all other power sources except USB. Official Linux support for Qgroundcontrol seems to be gone. It's just Mac & Win, unless you want to make your own port.
The PX4flow is indeed broken. The camera works. The altitude readings work. The flow compensated readings are garbage. Moving it around a lot gets useful flow compensated readings to start showing. It reverts to garbage after a certain amount of time below a certain amount of movement.
The gyros would have to be broken, but no mems gyro has ever died. Some of it could be the UART dropping characters. The raw readings are quite erratic around startup & engine arming time.
After a full day, the electronics were transferred to a new airframe with new motors & new propellers. The IMU drift, as it was known, was no more. So the motors should be considered gone after 1 crash.
There was 1 phase in the development where it was spontaneously getting thrusted into the floor by 2 motors going to full throttle. That might have done it.
The autopilot is now non functional. The PX4flow might be damaged. There was another reflex to put the sonar on foam, which probably doesn't work, either.
After a 2 day process, an outboard IMU was fabricated.
A magic switch was installed to switch the UART between debug & IMU input. The bootloader started in 115200, then the mane firmware switched to 230400 for the IMU. That supported 1050 full readouts per second.
Hard mounting the electronics made all 3 axes drift for the 1st time. So the next strategy was to get as many axes stabilized as possible, then use an offboard gyro for the 1 remaning axis.
Returning to nearly the original battery mounting still resulted in drift in 2 axes. Also tried the hardware lowpass filter setting & increasing the full scale range.
That left going back to a very large offboard IMU in a heavily padded case.
Finally added feedback for the rate of the rate gyros that all the commercial quad copters seem to have. They call rate of rate feedback "stabilization P", while the rate feedback is "stabilization I". The feedback for angle & trim are the "self leveling P & I". A total of 4 terms are required.
It didn't have any obvious impact. The weight of the PX4flow is degrading the flight characteristics.
So the pondering about cubesats was lame, but now there is a new idea. Get a cubesat to photograph a spy satellite up close. It can't be very far from happening.
The mane spy satellite of interest is USA 224. It's in a sun synchronous polar orbit that passes over Baghdad at noon. It's low enough for a cubesat to reach. It can only be seen from the ground in Australia. It is believed to be a shortened hubble space telescope which can be approximated by chopping off hubble's instrument section.
Another spy satellite worth visiting is USA-223, the largest satellite in orbit.
It has a 100m dish. It's in a geosynchronous orbit over the middle east. A cubesat would have a hard time reaching that orbit.
Was hoping Elysium would have some stunning visuals of a space colony, but all you get are 2 hours of Matt Damon bleeding & some lame immigration policy narration. Finally figured out why everyone suddenly got obsessed with immigration policy in 2013, manely the lack of healthcare entitlements to illegal immigrants.
As one who hasn't had any health insurance since 2009 & has only seen people leaving US to obtain healthcare overseas, this story never made any sense, but it is what the media tells us to think. These movies also forget that once the immigrants own all the stuff in Elysium, what do they do about all the citizens who now want their stuff?
The conversion of Jodie Foster & William Fichtner from unemployed astronomers to most evil pair in the universe is complete. Never seen so much evil.
For all the accusations between republicans & democrats, the immigration policy debate is more of a class war than a party debate. The richest 1% want cheaper labor. The poorest 1% want entitlements. The middle class wants to keep their jobs.
The leaders of the system want expanded immigration while the workers under the system want less immigration. The government & the people are at odds, as every attempt at government has been.
Whether to port bits of Arducopter to Marcy 2 or port bits of Marcy 2 to Arducopter has been the mane question. Arducopter is a very large & growing nest of features, with a lot of hype around a few features that might be useful. Porting bits of Arducopter to Marcy 2 has been the only thing that fits into the 2-4 weeks between all up demos.
The additions or at least the hype is getting more frequent, as 3D Robotics has more money to invest in features, but the hardware has remaned the same. Recently, the big ones have been autotune, hybrid mode, inertial navigation, & Kalman filtering.
As with everything nowadays, the marketing for every new feature is in your face & viral, while reality sometimes differs. Actually having a stock arducopter to test with optimal parts is financially prohibitive. $800 + stock RC system + crash repairs is quite a lot by Chinese standards.
The inertial navigation still seems a bit like smoke & mirrors. There hasn't been a side by side test of inertial navigation vs. GPS only. The question is how much new information they can actually get from the same old garbage input.
The autotune feature is potentially useful, conceptually simple, but very time consuming to perfect. It's in 1 small file called auto_tune.pde. It commands a bunch of rolls & pitches to optimize an existing set of tuned attitude hold gains. It doesn't touch heading or position.
Ran with the RC car being manually controlled. Keeping it on the narrow urban sidewalks was impossible. After leaving the urban jungle of Rain Ramon, it could start staying on the 15 ft wide trail. After several miles, banging out a straight line became automatic. Its tiny wheels made it erratic on the crumbling parts of the trail. 1 handed driving is still difficult, but better. The controller needs to get smaller.
The gopro naturally died after 6 miles. The RC car went 7 miles on 1.5Ah, then suddenly died & had to be carried 6 miles home. That was way beyond expectations. Thought it would go 1 mile. 6 miles on 1.5Ah should be considered the range. It has a brown out shutoff. A brushless motor should get it up to 9 miles. Its nominal pace was 9m30s.
This system is fast enough to be a pace maker for the slow, long run, but needs more NiMH batteries which are super expensive compared to Lipo. Would only invest further development in a high end brushless system. Those things are expensive, even from Hobbyking. They're $75 - $100 just for the frame.
Don't think the changes are substantial enough to sell a manually controlled, modified RC car as a workout coach. The only differences are the 1 handed control, longer range, & attachments for water.
Making it completely autonomous is pie in the sky, but there might be a way to get it to drive the straight sections or avoid obstacles. A camera which automatically tracks the runner would be something. There's absolutely nothing which can make a timelapse video of an athlete during an entire run, but it can't be far off.
Interesting to note how the speed followed the NiMH discharge curve for 7 miles.
Lion after 6 miles. There's no way to see yourself this far from home besides robot.
A better battery, proportional steering, variable throttle, autonomous athlete following, & it could be a useful exercise tool. The reality of crossing streets & staying on the path makes the autonomous following hard, but the rise self driving cars gives hope.
It's really crusty. The steering is binary. The speed is binary. Instead of a steering servo, it bangs a motor completely left or right to the stalling point. It can't go backwards.
For someone who never had an RC car before, the amount of mechanics required to perform steering & optimally transfer power to the road is quite amazing. 4 wheeled vehicles are pretty complex & there's no simpler way of doing it.
The 2 wheeled balancing robot eliminates a lot of the complexity, but it still can't achieve the ultimate fluidity in turning that fully articulated front wheels do.
A manually controlled vehicle might be viable for exercise. A vehicle which carried supplies for 30 miles has been on the radar for a while. The quest began to make single handed driving more ergonomical.
It's not necessarily quad copter, but tractor based. Did not know Calif*'s mane crop was lettuce or that it produced any vegetables at all, since there hasn't been any rain since 2012. Though quad copters harvesting lettuce is a highly dubious proposition, there's certainly money to be made from tractors.
That guy was rich. Listening to him complain about the lack of PhD's from Stanford to design computer vision algorithms, you thought just stop being a dick & hire people who know how to design computer vision algorithms instead of insisting they have a PhD from Berkeley. Most of the discovering, inventing, engineering, living & dying in this town is done by makers & hackers working in garages. Yet another overstuffed taco salad supported by the federal reserve.
Your other thought was agriculture might actually become a big business as our generation believed 20 years ago. Unlike what we believed, the money is not going to biologists who study genetics, but to designers of robots & businessmen.
There are no jobs for biology majors who learned about robotics on their own, only for engineers who learned about biology on their own or businessmen who can track down farmers with bucks.
It's probably a supply & demand issue. Few people can pull off the engineering side & business side while everyone can do the biology part.
A huge aversion has emerged to the genetic engineering we thought would comprise most of the expansion in food supply. A lot of that is lack of knowledge about genetic engineering & movies, but whatever. For now, you can't use genetic engineering to feed Africa. That leaves using automation to expand the biological system we already have.
The sight of the 6th Street bridge in a music video made it unbearable to not do some running. It's the bridge of despair & the bridge of memories.
There was great despair to cross paths with the Jesus Heroine twice on it, a tiny fraction of a half marathon apart. There was being eyeballed with great intensity on it & longing to be in that parallel race just a few hundred feet ahead. These memories will soon recall a much younger & distant age.
In 2015, it's going to be torn down & replaced by a seismically more robust & hideously ugly piece of warped steel. The asphalt where me & Jesus Heroine ran will be no more. The backdrop seen on those days will never be reproducible again. There will be no trace of what was.
So many other men have memories of that bridge, there's a website about it. http://www.the6thstreetbridge.com/ It's getting more & more common with increasing age, for our past to be erased as structures from our past wear out, become economically unfeasible, & get destroyed.
Clicking through some ads reveals there's now a flood of banging good GPS autopilots out there, being marketed towards RC pilots. They're a lot more stable than anything I could afford, owing to the latest GPS module, barometers, & the arducopter source code, which they all undoubtedly copy.
Gone are the days when someone would tear them open & study the part numbers. The new customers are RC pilots with no idea what's inside. It wouldn't be surprising to find uBlox-7 & MS5611 in all of them.
Even the mighty Horizon Hobby finally caved in & started selling a GPS autopilot, though marketing it exclusively as a return to home feature.
The trick with autopilots is once you have one, now what? They're good at making timelapse movies. They can fly to places that were impossible for a human to reach manually, but the newest FPV systems make those places accessible for a human.
They're supposed to be good at returning to home when contact is lost, but there's very little evidence of that being used. It's still a lot of trouble to wait for GPS to lock on, make sure everything is calibrated correctly, just for that 1 feature.
There was some energy behind the following camera concept, but they probably weren't accurate enough to do a good job. Now there's hope for massive agricultural demand.
Autopilot for an RC pilot can get pretty boring, once the novelty of seeing it hover wears off. Hobbyists who design robot algorithms still seem to be the mane customers.
It was suprising that the mane media avenues: tested, flite test, rcmodelreviews were completely devoid of the GPS autopilot boom, emphasizing indoor quads instead. Even the mighty black sheep became famous not because of an autopilot but manual control.