Jack Crossfire's blog View Details
Posted by Jack Crossfire | Today @ 01:18 AM | 503 Views
Lions could imagine a number of daring scenarios where robots released ballast from the platform, strung new cables, or released tension from the existing cables to avert a collapse. It turns out all these minor tweeks were explored by the government already.


The mane problem is so much of the cable tension was just from the platform's structure, no amount of de tensioning or dropping of counterweights would have had any significant effect. Adding a new 1" thick cable would have done nothing. Making any difference would require rebuilding the entire 1963 cable system from scratch, probably out of reach for any automation.

The best lions can think of is quad copters laying down a completely new cable system above the existing cable system, 1 strand at a time. This would provide enough redundancy to allow humans to go in & replace the original cables.

Some more interesting points are that the original cables had strands popping since 2003 & they were never replaced due to lack of money. A large batch of strands popped in the 2014 earthquake which were repaired. The rate of strand popping undoubtedly accelerated after Nov 6.

What was done in 1963 can never be repeated, manely building a new telescope. It would cost many billions in today's money. China's much loved FAST telescope is manely a facade.
It doesn't have a transmitter or any liquid cooling & it's limited to much lower frequencies. The 1963 telescope was partly for missile defense & the government's job has moved a long way from military & science projects.
Posted by Jack Crossfire | Yesterday @ 04:43 AM | 3,664 Views
Pretty disappointing RPM's with the large 70 turn motors. Built up a new board to allow direct control of all the enable pins, which got it up to 272rpm on 10V. It would need 35V to hit the required speed with no load, 40V for margin. The next step would be reducing the turns to 20 & increasing the current.

Trying the motors with a sensorless ESC got them up to a slightly faster RPM but much less torque. Standard ESC's can barely get them started on 12V but use the current a lot more efficiently. The size of the motor drastically increased the required voltage for a given number of turns by virtue of the resistance.

Over the years, lions have avoided cleaning flux whenever possible. It's manely necessary for adjustable regulators. It otherwise might improve water resistance.
Posted by Jack Crossfire | Nov 26, 2020 @ 08:06 PM | 2,996 Views
Today's lesson was the hall effect sensor needs to be a long way from the magnet to get good results as an angle sensor. 3mm was all the room in the motor mount. If it's too close, you get large areas where the sensor doesn't pick up any flux. If it's farther away, you get more flux in the dead areas. This might be caused by the flux lines returning to the motor shaft instead of the magnets. The same thing might not happen in the joysticks.

An algorithm for converting 2 sensor voltages to an angle is a buster. The sensors never return the same voltage twice, so it ends up being a pythagorean therum problem which has to be optimized. A conventional sensored speed controller is just concerned with whether the sensor is N or S, so it can use a simple lookup table.
Posted by Jack Crossfire | Nov 24, 2020 @ 08:25 PM | 4,284 Views
2 full days of soldering yielded the fully wired truck with mockup of the user interface panel. A mane power switch was deemed essential, given the number of servo & bluetooth radio glitches which require resets. The mane power switch doesn't cut off the 2nd battery, though. The 2nd battery feeds in through the accessory connector & the hope was to rarely have a 2nd battery.

The user interface panel was envisioned hanging off the cargo area, encapsulating all the LEDs, radios, & switches. It eventually should also contain the debug header. It was a complicated wiring job.
Posted by Jack Crossfire | Nov 22, 2020 @ 02:45 AM | 5,127 Views
In lieu of any suitable sensored motors, these SI7213 hall effect sensors were released in 2016, roughly 10 years after the lion kingdom took delivery of its last hall effect sensors, the mighty Allegro 13xx series.

The digital signals from the new ones are a pain in the mane to read, but lions can appreciate what a leap they are from the past. Inside this SOT package is an ADC & a full oscillator with no external components. An oscillator used to require external capacitors.

The datasheet says only 2 sensors are required to detect a 360 degree range of angles. This requires extremely precise alignment. The best lions could do was sense within 2 degrees in a perfectly controlled environment. There's a 4 degree deadband where both sensors are saturated, but enlarging the unsaturated range would create a new deadband where 1 sensor was outside the flux.

With the motor bouncing around & dirt, this arrangement would have a hard time getting within 5 degrees. The motor phase goes around 8 times for every encoder revolution. It's definitely going to need an unsensored mode as a backup. That unsensored mode will drive it as a stepper motor to avoid the need for more parts to detect back EMF.

Fortunately, all lions care about is the relative direction to apply torque, to get the motor started. The torque is going to be limited to 5 degree steps. The tires might absorb some of the stuttering & the lack of a transmission helps. The 2 sensor encoder may not be accurate enough to get a totally quiet start, in which case there will be a major operation to replace the board with 3 hall effect sensors.
Posted by Jack Crossfire | Nov 18, 2020 @ 04:04 AM | 3,143 Views
Typical for these boards, there is much point to point soldering & wire harnessing after the etching. We have an IMU header, BT header, 900Mhz header, debug header, steering servo header, 2 motor sensor headers, 15V input, 6V input, 2 traction motor headers, battery voltage sensor.

The mane tasks are detecting motor angle & controlling the traction motors.
Posted by Jack Crossfire | Nov 17, 2020 @ 02:31 AM | 3,490 Views
Steering came together with only minor trimming of plastic. A slight toe out needs to be corrected later. The pushrods were quite loose after threading the PLA. There aren't any replacement tires.

Anyways, how to 3D print a giant cargo container on an Ender 3 has been perplexing. The leading theory is to divide the segments along the isogrid contours. This reinforces in 2 directions but not thickness. To get the thickness reinforced, they need another feature.
Posted by Jack Crossfire | Nov 16, 2020 @ 02:43 AM | 5,648 Views
The original 4.75mm rod ends & 2.6mm metal rods are unobtainable, nowadays. Metric rods used to abound, 20 years ago, but the shift to home improvement has moved retail to imperial units. A new ball link capable of farstening a 6-32 threaded rod to a 4.75mm ball was needed.

PLA doesn't flex as much as ABS, so they can't be popped out for adjustment. Instead, the entire wheel will have to be unbolted & rotated however many turns.
Posted by Jack Crossfire | Nov 15, 2020 @ 12:23 AM | 7,373 Views
After much 3D printing, it was time to mock it up to make sure everything fit together. With a battery, motors, & servo, it actually, subjectively felt lighter than the lunchbox. The rear end looks horrendous, but it has a complex job of feeding wires for the headlights, radios, debugging, & reset out the back. It also has to seal in the battery. The mane problem is it feels flimsy. Most of the loads are supposed to go through the aluminum rods. It'll be driven to destruction.

It would benefit overall from larger features, 6mm thick isogrids, a larger battery hinge.
Posted by Jack Crossfire | Nov 12, 2020 @ 09:07 PM | 95,451 Views
The largest panel a lion will ever print emerged, manely the battery floor. It was dictated by the ender 3's size. Would never guess the entire bed size was useful. Just 18 layers took 6 hours. There is much under extrusion because in order to get a smooth 1.2mm thick plane, the top & bottom walls can only be 2 layers.

Water would go right through these panels. It's a question of how heavy & expensive they should be. The isogrids allow very strong panels with very little material.

In order to handle the increased weight of the direct drive system, the lion kingdom is going to have to invest $100 in flexible filament. The flexible filament tooling should allow 3D printed tires & a waterproof version of the paw controller. A different paw controller would have to be designed, since it would have to be bigger & with only buttons.

3D printed tires have been around & are simply an extruded cylinder with no sidewall. The inside of the tire is common infill which supports the tread. The springiness is adjusted with the infill density.

The motor wheel coupling has to be redesigned to bolt the tire in place instead of having the tire stretch over it. The tires have to be made narrower to increase efficiency. The tires still have to be springy enough to not need suspension.
Posted by Jack Crossfire | Nov 12, 2020 @ 04:15 AM | 96,384 Views
Tires got mounted & the model for the traction section took a leap, as the march continued towards custom truck. Isogrids greatly complicate the 3D printing. The model is such a beast, 3D printing is just going to generate many cubes which have to be glued together.

Accessing the motors when it's fully assembled is problematic. At least 10 farsteners have to be removed. At least taking out the transmission freed up a gigantic space for electronicals & a single battery. A 2nd battery would still have to go in the cargo area.

The tires went on a bit too easily. Shrunk the ring size on the wheels, which might have been a bad idea.

Direct drive motors are so much heavier than a gearbox, the bare motors cause the tires to sink a lot more. It's going to need either stiffer tires or some kind of stiffener in the existing tires. At minimum, they're going to wear out instantly. A completely quiet drive train is expensive. A belt drive with smaller motors would solve a lot of problems but be louder.

At least 3D printing timelapses moved forward. It needs longer bolts & washers if it is to last.
Posted by Jack Crossfire | Nov 10, 2020 @ 07:23 PM | 9,452 Views
Castellations have proven a good way to print cylindrical segments which are to be fused. The downward facing side has to be .4mm taller than the upward facing side to account for filament sag. There needs to be a large gap between the cubes & horizontal alignment structures to account for elephant's paw. The cubes can't extend any wider than anything that is to slide over them, like rings.

The largest outer diameter possible is 48.9 because of the tire size. Another .6mm is required for horizontal expansion. 4mm are required for strength. Finally a 0.8mm inset is required to align the rings, which limits the motor to 43.5mm.

The coupling has a middle & 2 side diameters where it grips the tire. It has rings to align the tire. The side diameters have to be wider than the middle diameter to align the rings. The middle diameter can't be wider than the tire size because the tire has to slide over it. It may be that making the middle diameter narrower than necessary would make it easier to get the tire on.
Posted by Jack Crossfire | Nov 10, 2020 @ 02:42 AM | 5,148 Views
After much grinding & 10 hours of battling through broken or tangled PLA filament. The PLA filament doesn't like being stretched for many days & the spool came tangled from matterhackers. It has to be completely unwound onto another spool.

It actually makes reasonable torque, but not as much as the gearbox. There was some effort to expose the motor to air.

This coupling can be redesigned to fit up to a 44mm motor inside the tire. The only limiting factor is the tire inner diameter of 48.9mm. The tire has to slide over the entire coupling, so no part can be wider than than 48.9mm.
Posted by Jack Crossfire | Nov 07, 2020 @ 03:37 AM | 9,105 Views
Lions have long been mystified by how the inner wheel turns more sharply than the outer wheel in a car. The magic of automotive steering is achieved by "Ackermann steering geometry", not having a right angle where the pushrods push the wheel shafts. The V angle of the wheel knuckle causes the pushrods to push around a bigger radius for the outer wheel & a smaller radius for the inner wheel. The V angle is determined by the distances between all 4 wheels . The steering knuckles from a lunchbox can only be used with the same wheel spacing of a lunchbox.

Steering linkages also have a caster angle to make the steering wheel self center when you release it. That's not a spring in the steering wheel but the center of pressure on the tires following behind the tilted axis of the steering linkages. The caster angle makes the steering heavier & requires power steering to overcome.

In a computer controlled vehicle, a caster angle is still required to overcome flexing in the steering linkages. The lunchbox had a 10 degree caster angle.

Caster angle was invented in 1896. The Ackermann steering geometry was invented in 1817. For the thousands of preceding years of roman chariots, horse drawn wagons, egyptian pyramid building, black plague, fiat money, slavery, wars on disinformation, conspiracy theories, & collapsed governments, all the world had was brute force, skid steering.
Posted by Jack Crossfire | Nov 04, 2020 @ 06:26 PM | 6,760 Views
After 18 months & not many miles, the motor died. Its bearing failed some time ago, leaving the shaft to rattle inside & make all kinds of noise. It accumulated dirt, eventually smashing a capacitor, causing the hall effect sensors to become intermittent & the tachometer to fail. Thus, the vehicle was dead. Its spare parts will be put into a new vehicle over the next year, while the 1st vehicle comes back from the day job to fill in.

The autopsy revealed the hall effect sensors were directly against the rotor magnet. There was no 2nd set of magnets for the sensor. The rotor magnet was a single cylindrical magnet with grains magnetized to create alternating, radial poles.

Original photos reveal a snug fit in the original bearing. It must have worn down. These motors are still sold as "High Efficiency 540 17.5T 2200KV" under various shell brands but the task of incorporating a sensor just requires a flimsier bearing.
Posted by Jack Crossfire | Nov 01, 2020 @ 06:44 PM | 6,405 Views
The answer is no, sensored brushless motors don't sense the rotor magnets with hall effect sensors like lions did 8 years ago. They use a separate encoder outside the rotor, with its own magnets. There are rare, oblique views of an encoder, with 3 hall effect sensors.

How To RC | How To Rebuild/Clean 1/10 Sensored Motor (7 min 51 sec)

For smooth, quiet starts, you need a continuous 360 degree angle measurement. The datasheet says you can get it with 2 hall effect sensors & 2 magnets. Digikey is like a box of chocolates. These hall effect sensors ended up being open drain digital. They spit out 1000 readings per second using pulse duration encoding with 5 usec precision so simply testing it requires building all the electronicals.

Based on a historic wheel spacing of 140mm, there's still enough space to have a battery, motor encoders, & motor supports in the space that was formerly used by a transmission.
Posted by Jack Crossfire | Oct 31, 2020 @ 09:52 PM | 7,385 Views
These are the 1st motors specifically bought for direct driving wheels. All previous motors were bought for quad copters or brushless gimbals. The journey begins with stock Propdrive 42x48mm 650kv motors, $30 each if they're lucky enough to be in stock. Rip off the C ring & write the numbers on all the poles.

The hardest part is removing the stock winding. It was easier to remove from Hextronic DT750's. It just came off in a continuous string. The Propdrives are so heavily eurethaned, the only way to get the windings off is to cut down the middle without damaging the powder coat. Then rip off the windings from each side of the pole.

All the leftover eurethane must be scraped off the poles to free up enough room for the maximum turns, yet without damaging the powder coat. The lion kingdom uses the same winding diagram it used 8 years ago from an unknown internet forum.

To get the most torque, put on the most turns that will fit, but more turns also reduce the maximum RPM. The lion kingdom guessed 70 turns of 32AWG, but after finishing, the motor would clearly fit 80 with careful winding. This resulted in lots of torque & heat at 10V 500mA. This was the hottest lions would take it with PLA & it would be for starting only.

Lions don't really know the required torque or have a way to measure it. Going 6 miles with maximum torque applied to 2 motors would burn 200mAh/mile, such a ridiculously low number, lions suspect the torque is too low, the...Continue Reading
Posted by Jack Crossfire | Oct 30, 2020 @ 04:13 PM | 7,654 Views
Given the 3 days required to fabricate, it's never going to be mass produced. The wires should all be wrapping wire. They would have to be soldered on a separate jig, then glued in place.

The hall effect sensors were off center. The screw & the speaker need to change places. It needs a stand to balance on the charger. At least the frequency hopping works & uses a very simple algorithm. The hall effect sensors have to be the lowest mV/G possible.

Got the hall effect sensors as close as possible without removing material. Spacing out the magnets more just increased the deadband. The hall effect sensor fiducials are exactly where they need to be. They could be shifted to allow for hot glue residue. The magnets could be closer to the shaft, but that would require going to a metal shaft. There's still a bee's dick of hot glue residue offsetting the steering sensor. Hot glue residue could be a legitimate reason for adding clearance.

Only a driving test would show if the hall effect sensors are good enough. Getting within 0x08 might be good enough. Gluing the hall effect sensors has been a matter of dropping a blob on the plastic, then pressing in the sensor with the mark 1 lion paw. It cools the hot glue instantly but might be the most accurate way.
Posted by Jack Crossfire | Oct 28, 2020 @ 06:28 PM | 9,057 Views
Obviously a lot more than a few tweeks happened. The mane event was moving the speed buttons to the sides & the power button to the back. The button spacing was dictated by mechanical clearance rather than ergonomics.

This would be a lot easier with tact buttons, but the lion kingdom insisted on something more durable. These MHPS buttons could have gone down to 6mm per side without a major increase in cost, but the lion kingdom wanted to finish. Despite every effort, hot gluing all the hot glued bits is still a buster. What's needed is a fine pointed hot glue gun.

There's some debate about whether the pinky grip should go all the way across. The 2 halves don't line up, making it look awful.

Too bad there isn't any more investment in lion machine interfaces. They all standardized on the game pad, 20 years ago. There were a few voice, gesture startups, 10 years ago.
Posted by Jack Crossfire | Oct 26, 2020 @ 03:26 PM | 7,037 Views
The 1st prototype had no showstoppers, just a few tweeks. The speed control buttons would be easier to reach if they were side by side. This would allow the power button to go in back. The height transition was messier than hoped. Throttle needs to rotate 30deg north. Pinky hook should extend the full height. Speaker hole could be smaller. It's heading to a dogbone.