Jack Crossfire's blog View Details
Posted by Jack Crossfire | Today @ 06:54 PM | 405 Views
With no payload, it managed the required 10mph on level ground, but slowed way down on a slight uphill. There are hillier routes for more testing. With a 2lb payload, it managed the required 6.66mph on the standard road grades. The 6.7 mile effort burned 280mAh/mile, more typical for the lunchbox. PID adjustment for throttle might be in need. Getting more performance numbers requires implementing the speed paddles, which requires dealing with android studio.

Some TPU bumpers were a desperately needed safety feature. Angle aluminum ends at 10mph are lethal. Even better would be some cubertruck front end, but lions only have black TPU. Maybe a PLA cybertruck front end could be adhered to TPU cushions. Lions bash it into curbs all the time, so TPU is a better option.

Lions are keeping an eye open for a cheap pancake motor. These motors were selected for the lunchbox tire diameter, but wider motors with thinner tires would be more ideal.
Posted by Jack Crossfire | Yesterday @ 07:41 PM | 3,667 Views
10.3 miles burned 217mAh/mile. The last mile was with a heavy payload. This exposed a serious lack of torque, compared to the lunchbox. Integrals tend to wind up & send it flying on level ground. It can't get up any hills. The motors didn't obviously get hot. There's definitely a power band as evidenced by the integral windup.

The lion kingdom is undecided on what to do about the paw controller. Making just a mockup with those intricate controls takes forever. There's some debate about getting rid of proportional steering when it's stopped & even more, getting rid of proportional throttle. The hall effect throttle would just be for detecting reverse.
Posted by Jack Crossfire | Jan 15, 2021 @ 07:19 PM | 6,696 Views
Version 3 finally managed to go 6.4 miles with no payload, over hills & bumps, with minimal stops. Power consumption was a whopping 175mAh/mile. It could go over 20 miles without a payload, on a single battery. It could also go 10 miles on a much lighter battery. The motors might benefit from a hotter winding, pending speed tests & payload tests. The motors weren't obviously hot after the drive.

Despite the bumps, the encoders stayed in place. The PID controllers were soft. The hard tires made it squirly. The mane limitation was now the paw controller. The controls need to be spaced out 10mm in every direction & it needs to be bigger. The lion kingdom has been leaning towards getting rid of the isogrids & the dual thickness in favor of simplifying it.

The battery compartment needs foam. The bolts need lock nuts. The tires held up well. Despite the hall effect joysticks, binary steering proved much easier than analog steering.

The low power consumption was manely from the hard tires & very little contact patch. Less benefit was from being direct drive, lions believe.

The encoders still required printing spacers to tweek the alignment. They're not as stable as hoped, but at least they just need 1 magnet & 2 sensors. Having the sensors too close to the magnet causes bigger constant regions of voltage. Moving the sensors farther from the magnet axially creates more changing regions. There still might be a way to make the software use past encoder readings to deglitch the current reading.

The steering PID controller needs to reverse direction when driving in reverse. The lion kingdom only realized it after 7 years of failing to hold a straight line in reverse. The trick is transitioning between fwd & rev.
Posted by Jack Crossfire | Jan 14, 2021 @ 02:46 AM | 5,100 Views
A batch of heroic soldering, a reversed buck converter, & the motors came to life. Mounting those right angle hall effect sensors took some doing. Mounting the motors was another buster. Things aren't as modular as hoped.

A few dozen more farsteners would go a long way. Given how many farsteners were in the standard model copter, the lion kingdom might have over emphasized efficiency of farsteners. The wires aren't very vibration proof either. It could stand to use another plate with wire traps .

After voltage testing, a few strange sounds, the wheels lunged forward. It was still running firmware for calibrating the motors & it showed the drivers were installed correctly.
Posted by Jack Crossfire | Jan 13, 2021 @ 12:41 AM | 3,413 Views
After a moment of silence before sacrificing $30 of hardware, the decision was made to bend all the pins in reverse, grind holes in the enclosure, & bolt the L6234's on the angle aluminum with their heat sink sides out. It required 22 new jumper wires to put the L6234's on breakout boards. It now is quite packed, despite containing only 1/2 the total amount of wiring.
Posted by Jack Crossfire | Jan 10, 2021 @ 04:14 AM | 2,254 Views
It got all the way to the SPI connection being wired up when some more tests revealed the mane loop only ran at 15khz with the mosfets firing at either 8khz or 16khz. Lions believe it would need to get at least 100000 bytes/sec through SPI to drive the motors from the ARM. The hall effect sensors have 20khz bandwidth & each mosfet write would need 5 bytes. The 5 bytes would be a start code, power & phase for each ESC. There's not enough horsepower in the ESC to convert the hall effect sensors to a phase.

So it was back to improving the cooling for the L6234. It's actually rated for 4A but hard to cool. They recommend via stitching to a ground plane with heat sink on the back, but that adds thermal resistance. It could be the 1st time the lion kingdom paid for a manufactured board. The most efficient cooling is from mounting it upside down, covering the pins in captain tape. A heatsink could be bolted directly on it. That would take precise milling of the board or stacking 2 separate boards on each side of it. Quite a big assembly but maybe smaller than an ESC. Another way is just to lift all the pins so the chip can be soldered upside down. There's also doubling up the chips.
Posted by Jack Crossfire | Jan 08, 2021 @ 03:23 AM | 3,147 Views
The easiest way to drive sensored, direct drive motors is to drop in an e-bike motor driver. Those are much cheaper than hobby ESC's in price per watt, but they're gigantic & manely require over 24V. They also require 3 hall effect sensors instead of 2, sensing the actual rotor. They might be shrinkable by taking out the guts. This is quite an improvement over 10 years ago when there were no e-bikes & large motor controllers would have all been super expensive.

Given the limitations, the next option is hacking hobby ESC's. The supersimples from 15 years ago are rated for 35A but always burned out above 5A. Full power requires current & voltage to be in phase, but they're worth trying. They have no hardware UART exposed. Lions 4 years ago wrote ESC firmware with a software UART driver for debugging at 9600 baud. It's always depressing to reuse ancient code because it shows how clueless you currently are about something you found easy long ago.

The plan is to send commands through high speed SPI & have them both on the same bus. Programming them requires unplugging the STM32 & plugging in the arduino. The exact wiring for the least amount of unplugging is a problem. It almost makes sense to make a bootloader program them over 9600 baud UART, but lions expect them to quickly burn out & be replaced by a home made motor driver.

At least a mockup of the existing pieces revealed a lot of empty space. The glued corners are truly a mess.
Posted by Jack Crossfire | Jan 07, 2021 @ 03:14 AM | 23,836 Views
Another incremental improvement. It needs a piece spreading the struts on top, but is otherwise useful. It currently uses a bend in the wheel forks to adjust height. It would be stronger if the wheel forks were straight & it adjusted the strut height, with the struts having a spreader on top. This reduces ground clearance, though.

The rod ends split if they're tightened with a socket wrench. They have to be tightened with pliers.

The battery compartment could have more ground clearance if it had bends like the steering section. This would require gluing a lot more pieces.

After ordering another $40 of PLA, it became clear that perfection really requires a way to recycle PLA. Every model is otherwise going to end 90% of the way there.
Posted by Jack Crossfire | Jan 06, 2021 @ 01:25 AM | 28,644 Views
The next steering module looked like a darth vader respirator. The honeymoon with isogrids started to wear off. It's been a time consuming exercise. More parts are just using infilled cubes because they're much faster to model. Infill is the same as isogrids, but hidden. A test panel uses 7.4m for an isogrid & 9.5m for a 25% infilled cube with the same wall thickness.

Traction module was disappointing to glue. Gluing right angles was a failure. Right angles have to be printed into the part as before. Made the bathtub a single glued part, with temporary farsteners.

The last lunchbox had a tire change which turned into another major repair. Silicone caulking didn't survive on tires, so it was back to waterproof sealant & another lean towards ordering more tires. Surprised how worn down the ABS parts became, after 5 years. Even parts which don't get bashed are wearing down from years of slight rubbing.
Posted by Jack Crossfire | Jan 05, 2021 @ 12:24 AM | 2,960 Views
24 hours of printing with manely the 1mm nozzle yielded the next traction module. Being 100% infill & 20mm wider made it a heavy monster. The 1mm nozzle with 100% infill actually makes an acceptable pattern on the top layer, so 1.2mm thick panels should be reduced to 1mm.

A few panels have to be reprinted. Layer heights were really bad. Some 3.6mm panels came out only 3.2-3.4mm thick. 4mm panels came out 3.7-3.8mm thick. 2mm panels came out 1.7mm thick. Horizontal expansion remaned while layer compression abounded.

Since lions manely print panels, the thought has occurred of trying to make the panels out of ABS. The panels transfer most of the force. The parts which risk warping aren't transferring a lot of force, so they could stay PLA.

This module uses a lot more farsteners than the lunchbox. The lunchbox had a bathtub forged in a single piece. The thought has occurred of printing farsteners & gluing them in place, just to reduce the number of farsteners. With the motors modularized, the traction module can now be glued with only 2 removable panels. 10 farsteners could just be temporarily installed for gluing.

The steering module is a buster.
Posted by Jack Crossfire | Jan 04, 2021 @ 01:20 AM | 4,375 Views
After much modeling of a new traction module, the new motor panel came off the print bed. Used some TPU pads to try to make it compliant, but with 2 bolts, there is no compliance. They would need 4 shorter bolts to have any sliding motion. Then, the problem would be keeping the TPU pads from slipping out of alignment.

The new traction module is 20mm wider, on account of the narrower tires. The tires still have 20mm to go before they're as wide as the original lunchbox. It should have a lot more room for the electronicals & be stronger. The steering module is up for another revision. Lions have settled into a workflow that uses sketches for most everything instead of defining primitives for every hole or outline.

Lions realized the entire traction module could be printed as a single piece of plastic, with supports & infill instead of isogrids, tabs & slots. The isogrids, tabs & slots make it stronger & reduce the cost.
Posted by Jack Crossfire | Jan 03, 2021 @ 03:44 AM | 4,516 Views
It wouldn't have been a very productive day if it didn't yield new pliers. Helas, they don't lock but they do have a fair bit of friction. Locking just requires more adjustments in clearances & heights. There might be a better way to lock them with a hook instead of a ratchet. Even better would be plastic farsteners for the locking ratchet instead of glue, but these are expensive. Printing .4mm line width with ironing gave the desired waterproofing & tooth precision. Going without supports made it stronger & reduced the print time.

Freecad is a bit like chizzling granite, but let's consider 25 years ago there was basically no free CAD program & the options were very expensive Windows programs. The one they used in school was Pro Engineer, now called Creo. It only ran on $10,000 UNIX boxes. Computing requirements by the standards of 25 years ago were quite intense.

Most of the crap made in the 1990's would have started life inside UNIX boxes. Before then, lions believe crap started life with manually operated machine tools & paper drawings. More complicated crap like airplanes might have started life inside a VAX, at great cost. The electronics in the Gallileo probe were designed by paw in the 1980's, so consumer junk would have all been modeled by paw, all the way until the 90's. A lion with freecad & a 3D printer today is like a god at the head of a large team of draftsmen & machinists in 1980.
Posted by Jack Crossfire | Jan 02, 2021 @ 01:49 AM | 6,759 Views
2020's biggest contribution may go down as discovering functional 3D modeling. Before September, lions had only done artistic models for games. Would consider it a pretty big gain in knowledge from those 1st steps to today. Lions had virtually no 3D modeling skills until 2018 & now might just break even with today's high school graduates.

It's understandable that 3D modeling wasn't taught in school 30 years ago, but nowadays it's all but essential. It's an example of how old age can make you obsolete. By the end of public school, lion classrooms had Apple IIgs's, crippled to 2.8Mhz. They couldn't teach anything close to functional CAD modelling.

The radial motor encoder proved acceptable in some tests. Its mane advantage is more stable alignment. Getting it required revising the PC board pliers. They were revised to more closely resemble the lion kingdom's beloved 30 year old radioshack forceps & resist breakage. Helas, the radio shack handle uses a single part while the lion handle uses 2 different parts, on account of the need for a surface for clamping.
Posted by Jack Crossfire | Jan 01, 2021 @ 03:20 AM | 7,819 Views
A 3rd silicone bead went down. The 1st silicone beads went down in July. They seem to last 2 months. After wearing down 2 applications of silicone beads, they seem to do reasonably well at extending tire life, are easy to apply & affordable.

The robot is burning 386mAh/mile, which is almost as bad as a brushed motor. It could be higher friction with the silicone beads in addition to its 6 year old suspension. Ordering another set of tires is still possible.

Reviewing the lion kingdom's stash of 20 year old shaft collars, the thing to do is keep the encoder magnets directly attached to the motor shaft & install a collar to keep the shaft from sliding. To make the encoder more reliable in general, there was a new idea to have a single radial pointing magnet with the hall effect sensors pointing radially. This might be more resistant to stray magnetic fields & be more compact.

There was a desire to use TPU to give the motor some compliance, but it requires shrinking the encoder & using a lot more farsteners.
Posted by Jack Crossfire | Dec 30, 2020 @ 07:23 PM | 7,165 Views
This one got .5 miles before the motor controllers overheated. Time to repurpose ESC's as H bridges. Since they don't have thermal shutdowns, they'll need a timeout when full power is applied or no commutation is detected for a while. The ESC will have a timeout for loss of control signal. The confuser will have a higher level timeout for stalls.

The encoder magnets shifted closer to the hall effect sensors, causing the right wheel to completely die & the left wheel to spin more roughly. The motor shafts obviously shift in & out when they hit bumps.

The easiest but expensive solution is new C rings for the motors, but having the encoder integral to the shaft is a nightmare for taking the motors off.

Now that the tires are much narrower, enough of the rotors are exposed to mount the hall effect sensors outside the rotors. The motors would be a lot more modular. This would require 3 hall effect sensors per motor instead of 2. Another way is to have a gearbox or clutch connect the motor shaft to a self contained encoder. It would be noisier, take more space, but still make the motors more modular.

A lot of mane hair accumulated in a motor. The motors need a big shroud to keep dirt out, much easier with narrow tires.

After living with it for a while, proportional steering is a decided failure. There was no way to keep a straight line with the user directly controlling the servo & heading hold switching in only when the joystick centered. The controller has to go back to binary steering when driving & proportional steering when parked.

It held up to some light bashing. Only the front battery panel broke when the battery smashed into it. Structural changes can't compete with the motor issues.

The good news was it had plenty of torque & was quiet. The TPU axles were just stiff enough to keep the wheels from scraping. Had to turn off the music to listen for motor defects.
Posted by Jack Crossfire | Dec 29, 2020 @ 09:55 PM | 8,643 Views
The new front end was made entirely with the 1mm nozzle printing .8mm line width. TPU sections definitely get softer when they're attached to wheels. The current nozzle yielded smoother top surfaces when there was no infill.

The amount of TPU flexing definitely wasn't desirable. Diagonal struts outside the angle rod, joining closer to the wheels, would be the next step. They would allow the wheels to be spaced farther apart. The TPU axles would join to a big PLA triangle which formed the shield & joined the steering section to the battery compartment.
Posted by Jack Crossfire | Dec 29, 2020 @ 03:48 AM | 6,996 Views
The next great wheel mount. (0 min 37 sec)

The lion kingdom has converged on the compound tool for modularizing models. It's better than the clone tool, the part tool, the fusion tool, or the rigid body tool. It makes the complicated angles for ackerman steering much easier. So far, it's behaved like a boolean tool, but with the ability to have only 1 object.

There was also the discovery of transparency in the view panel under shape color. That made it immensely easier to align objects.

After the rock hard tires, the 100% TPU suspension had an unexpected increase in softness. TPU definitely amplifies the effect of structure. Even the cheap TPU can be very soft. It might work as a shock absorber, but some would have to be PLA. CA glue didn't adhere to TPU at all.

A mix of TPU & PLA gave slightly more practical suspension. TPU may end up too soft for any suspension. Currently, the PLA member experiences a twisting force from the TPU member.

The next step would be to join the 2 TPU bars in the middle via another segment, so the PLA members can freely pivot. To resist front to back force, the PLA members should have a triangle plate below the servo, that joins them to either the battery compartment or the opposing angle rods.
Posted by Jack Crossfire | Dec 28, 2020 @ 02:19 AM | 7,355 Views
So the battery compartment broke, too. Seemed to be the CA glue giving way. This problem was going to be solved either way with a tongue & groove, but printing it requires revising the motor mount.

The $26 TPU lions used for the tires is a very hard variant, like most $26 TPU. Real tires, more than in the academic sense, would still require very expensive filament & cost far more than lunchbox tires, not to mention getting the Ender 3 to actually work. The cheap TPU is still much stronger than PLA & the favored load bearing material. At least the rear hubs looked good.

Lions stare at the CAD models endlessly, clueless on how to get the most optimum transfer of force from the wheels to the frame, & wonder how CAD modelers can get paid so little compared to programmers. So much stuff begins as a CAD model instead of a piece of software.
Posted by Jack Crossfire | Dec 26, 2020 @ 06:20 PM | 10,612 Views
After 1 week of printing & $11 of filament, the container was manely finished except for the handles. It's a brick.

It contains 516g of filament or 1.14lbs. The coroplastic container was much lighter & stronger.
Posted by Jack Crossfire | Dec 26, 2020 @ 05:48 AM | 10,540 Views
The lion kingdom finally realized the reason 3D printed cars all have sections of black or are entirely black is because they're made of TPU. No-one gets enough strength out of PLA for the wheel forks.

There was still the dream of getting a lower profile than the standard RC car, via a rigid suspension, without losing vertical clearance.