A machine vision algorithm emerged, which could do a reasonable job tracking a path. The machine vision algorithm of choice goes at 3fps on the raspberry pi. The PI can only go up to 900Mhz. The RTL8188 is the only chip still supporting wifi access on it & it rarely stays connected longer than a few seconds.
The other problem is the algorithm only works in daylight, on a single uncrowded path near the apartment. When you have to commute 3 hours a day, you can't be home during the daytime, at all. There wasn't a chance to test it in January & probably won't be until April. There's no open space in the city to test it.
In terms of cost per clockcycle, the Odroid C1 has now been the duck's guts for several years. Development of the RTL8188 driver has continued on the Odroid, supposedly making it more bearable. It begins at $37, then the gauging begins in ernest, with exorbitant shipping, taxes, & cash only.
It makes no sense for so many middle men to now be required in payments, each adding on their own 2.5%. 30 years ago, there was just a bank. Now, it takes Paypal, Square, Apple pay, Google wallet, Coin, thousands & thousands of middle men. As middle men gauge more, new middle men emerge, promising to reduce the fees by spreading the cost of the lower middle men among a larger group. It was once said that Visa's exorbitant fees bordered on racketeering. Now, the current generation is in middle man euphoria, praising Google wallet, Apple Pay, & Square, which merely tack on other fees on top of Paypal, on top of Visa, on top of the bank. It's insane.
It's amazing all those formerly penny chinese businesses of taxi cabs, pet furniture, LED blinkers, & bike delivery are now multi billion dollar Google acquisitions. Unlike 1999, there are no IPO's. All the money is in buyouts paid to the founders. It takes a lot less people to do what took a staff, 15 years ago. There are no more tales of Netscape staff members retiring. No more tales of six figure secretaries. There are no more secretaries or IT staff. The whole show is just a CEO & an iPhone.
The long time quest for a KH-11 model ended when I remembered paper hubble models abounded. The KH-11 was once believed to be a short hubble with no instrument package. A later Kiwipedia update showed it being identical to hubble & having a large propellant tank for reboosts.
Decided to stick with the earlier theory. No matter what it really is, it's still vastly different than hubble.
The paper model wasn't as inspiring as hoped. It had to be 2.66x smaller than the NASA model to fit in the levitator. The toner fell off. The model was intended for a color printer. Left out the high gain antennas because they wouldn't fit in the levitator. Left out the instrument boxes because they would be crushed by the magnet & were microscopic.
BART has constant, massive delays. What is a 45 minute ride on paper is normally 1 hour. The delays are 1/4 caused by equipment failures & 3/4 caused by people. There is a big difference between the news & how happy people actually are, because they're obviously not getting by while the news would have you believe things suddenly got rosy on Jan 21, 2009. The reality is there are so many protests, crime, & health problems among people who are doing a lot worse than 6 years ago that nothing is working.
Living on a train for 10 hours/week is more productive than operating the break pedal in a car, but it doesn't leave any time to do anything but what can be done on a laptop. The sum total of a week of physical fabrication amounts to a very minimal revision of the Tamiya circuit board
The new board probably ended up being unnecessary, since it ended up manely removing most of the reworks instead of integrating them.
Current sensing - a decided failure. It was easier to set a constant RPM than try to adapt the RPM based on power. Couldn't get an accurate reading of current usage. The user can manually adjust RPM based on steepness.
Headlight MOSFET - not worth the extra board space. Only set once per drive. Required a gyro recalibration.
3.3V - needed connections for external regulator.
Servo - needed connections for external regulator.
ESC - Needed PWM signal on 5V pin with pad to jump to 5V. The pad didn't make it.
Last year, a car which followed the athlete instead of the path began to emerge as the best solution, but a car in front doesn't know where you're heading, only which way to turn to keep you in frame. This causes it to drive in circles.
Setting the car to follow a desired magnetic heading won't work on its own, because the magnetic heading isn't precise enough. The latest theory is if the car senses both a desired magnetic heading & the direction towards the human, it can stay on the path. This works only if the human is directly behind the car, with decreasing accuracy as the human moves alongside the car.
The car maneuvers so the angle from the human to the car to the desired magnetic heading is 180. The desired magnetic heading is changed from the stick controller to steer the car.
If the desired magnetic heading is off, following it leads the car off the path, but the human stays on the path. As the car heads off the path, the human-car-magnetic heading shrinks & the car turns back towards the path to make it approach 180 again. Wherever the human goes on the path, the car maneuvers to stay in front.
The 2nd case is the human alongside the car. This requires a different algorithm that maintains a fixed distance to the human. If the human gets farther away, the car steers to reduce the distance. This would be much less accurate than following behind the car.
The 3rd case is the human in front of the car. It just needs the...Continue Reading
Drove another 8.5 miles with the forward looking webcam. This would be the actual video technology in a path following solution. Had both batteries onboard. The old battery only went 3 miles & the new battery had more than 5.5 miles. Time to send Hobbyking another cash infusion. Also suspect 10min/mile is much less efficient for the current gearing.
Tried driving off a curb, but flipped, snapped off the wide angle lens, & this made the pi crash. Fortunately, it got 8 miles of footage. Forward looking video had much higher compression.
Edge detection showed promise, with gopro footage. Helas, JPEG compression erased too much detail & created too many macroblock lines. Would need to try again with the webcam in uncompressed mode. The wide angle lens was a waste of time for forward vision & created too many reflections.
It takes about a day to process & upload a video to the goog, so many miles of driving video have now accumulated, all on the same test trail. It's accumulated over 50 miles, since its arrival 1 month ago. To drive an RC car 50 miles, you have to run 50 miles.
A straight test of RPM vs PWM showed the dreaded stair stepping.
A test of minutes/mile using the most precise, slowest RPM feedback still showed stair stepping above RPM's where it oscillated. So either the Tamiya ESC wasn't precise enough or the mechanics had some voodoo. It's probably a limitation of all ESCs whether brushed or brushless. The next solution was to install the H-bridge from the G-buggy.
Whacked on the ages old 5V BEC to try to maximize the range, then tried constant RPM set to 500 or 9m18s/mile.
The result was higher speed going uphill than downhill, with the total speed in the 10minute/mile range. It must have been the lousy tracking of rpm.
Then of course, the range was reduced to 4.5 miles. The BEC was actually less efficient than straight PWM. As for measuring power independent of voltage, only the current feeding the BEC was measured, so the power would vary as the ESC efficiency changed at different voltages. The rough figures showed 15W on the downhill & 17W on the uphill.
Then, there were the usual software problems. Throttle always reset during fast turns & sometimes reset during slow turns. Acceleration was too slow. A faster tachometer is needed.
Ran all 9 miles looking at data on the phone. Probably need to capture the bluetooth data. During these long runs, a wish list topped by automatic path following always forms. Automatic path following can be reduced to a simple problem: determining what's 1 material & what's another material in an image. 1 material is usually asphalt. The other material is grass, gravel, or dirt.
The computer always knows asphalt is on the left & other stuff is on the right. It can get the color on the leftmost & rightmost parts of the image. Then it can work inwards until the colors change.
So what happens when power is constant & voltage is changed?
Was hoping there would be an identical percentage decrease in RPM for all power. Instead, the RPM for 25W decreased 9% while the RPM for 10W decreased 4% as the voltage increased.
RPM decrease at 12V vs 9V:
It was a table just as complex as scaling PWM based on voltage. The rainy season was over & there was enough time before the next commute to try out constant power at a roughly constant voltage anyways.
The speed variation on hills was wider than any other method. It became clear that humans need a lot less power going uphill & a lot more power going downhill than a motor to achieve the same speed. It's the same as how humans don't hear all frequencies equally.
The next step would be a hall effect sensor tachometer. It's all purely speculation about what a human can keep up with, faking a speed that feels like constant effort instead of trying to predict constant effort from power. The tachometer could be combined with an accelerometer to give it some variability based on incline. The current sensor is useless.
After much playing with LTspice, it became clear that it was erratic with small voltages. Sometimes it would actually work. Usually, it would go to full maximum. With large voltages, it always correctly simulated the difference amplifier.
It was time to build up a circuit with an LM324. Amazingly, it worked. It measured 1-5A nearly linearly, using a length of wire & some spare trim pots.
Using a high rail voltage with voltage divider cut the op-amp dropout voltage by half, making the LM324 reasonable. The bog standard difference amplifier did exactly what a proper INA169 sensor did.
With all the effort in the dynamo & the current sensor, you might as well try making the constant power throttle regulator you always dreamed of. Even with a power factor correcting cap, the current was still oscillating wildly from the PWM. Averaging the current & voltage down to 10Hz made a very stable power reading. It was an actual power reading from a bunch of spare parts.
Building a dynamo was long dreaded, but necessary to have any hope of regulating the speed.
Out of sheer luck, the Tamiya came with a tool which perfectly coupled 1 shaft end with a spare motor. A simple jig could hold everything together. The bench supply couldn't provide stable voltage because the current & voltage were out of phase. The old LM317 system with giant capacitor could provide stable voltage.
The load could be adjusted by shorting out different motor leads with different lengths of wire. An op-amp measured the current between 2 motor leads to give RPM. It was surprising that the op-amp was so sensitive, it could detect the current in a short piece of wire.
An input had to be grounded for this to work. Letting the motor leads float above ground only showed 0V. There was a lot of ringing when the motor leads weren't shorted, too.
Relying on the phone app to show voltage & the oscilloscope to show RPM. It had 1 more problem: the speed fluctuated. If it started slow & ramped up, the speed would be higher than if it started fast. Slight variations in wheel pressure also affected speed.
A software program would have to step though the entire PWM range for each voltage to build a table of RPMs. Then, given a target PWM value & target voltage, it could look up a modified PWM value for the real voltage that would give the same power. There are still dreams of measuring the current accurately enough to make constant power feedback.
So oil prices started plummeting at the same time quantitative easing ended. The media promoted it as a victory for Obama's energy policies & supply catching up to demand, but if that was true, prices would have fallen back in 2008.
More likely, the decline in housing prices that was averted in 2008 by printing money has resumed in a new form. Too few people are making too little money to sustain $100 oil. The government guarantees all mortgages against all losses, so $100 million bungalows can thrive even in a world with no jobs, but it doesn't guarantee oil. Even in Calif*, gas fell below $3 for the 1st time in 10 years despite a doubling in taxes.
So assuming it is the lack of money & not the reincarnation of Einstein you elected, & the government isn't going to subsidize oil prices with another $5 trillion, lots of people in N Dakota, Canadia, & Alaska, are going to lose their jobs. What's going to happen to Tesla when the allure of avoiding those scorching $3 gallons of gas is gone or are people buying Teslas because of something besides the fact that they're electric?
The promise of gas disappearing caused huge investments in lithium production for electric cars & a plummeting in lipo prices. To be clear, Teslas never used lipos. They used traditional, higher capacity ion batteries in metal cans. Still, they did create a lot of speculation in lithium production that benefited everyone. The world reinvented itself for a future without gas, heavily dependent on electricity storage.
The millions of people extracting oil & producing electric cars in a world with plummeting gas prices now look a lot like the millions of people working at webvan in a world with plummeting demand for mail order groceries, 15 years ago, but there wasn't socialism back then. Socialism is a new beast. The government could subsidize Tesla & the gigafactory with $4 or $5 billion. Google could easily buy them out for $50 billion. Technically, $100,000 electric cars can thrive in a world with no jobs & no demand for electric cars.