TinyStar, the 2-Day Dual-Flight-Controller Scale-Model Octorotor

chorl2 Comments

That might be the most specific descriptors I’ve ever had to heap onto one build. It’s also kind of hard to explain how it got started… which I guess is true for some other stuff I do too. But as a reminder, here’s that mysterious picture that I posted a while back when I was working on Tinycopter:

Why does it have eight propellers!? That’s like, two quadrotors at the same time! And why is it so pointy?

First off, it’s a reasonably accurate scale model of a Cinestar 8. This was 100% intentional – not only am I out to one-up the quadrotor…uhhh, arms race? at MIT, but my buddy Shane Colton is way better at getting free stuff than me and he has a Cinestar 6 frame. But it is used for custom controller testing, so it has a legitimate purpose despite being free (an exercise of the classic industry strategy of “give academia free stuff so they can do weird things to it”). The Cinestars are probably some of the most badass-looking multirotor frames around – totally murdered-out and made of carbon fiber and black GFR nylon, and they kick ass. So much. Also, nice cameras make everything look more epic.

Second, the cool thing about Tinystar is that it uses two independent multirotor controllers (Hobbkying i86, which may be based on KKMulticopter’s firmware, but it’s just different enough that I’m not confident on it) feeding from the same receiver. It is quite literally a Siamese quadrotor twin – one controller is run in “X” mode and controls the 45 degree arms, and the other controls the 90-degree arms in “+” mode. This was a ‘dumb but potentially awesome hack’ idea spurred by discussion at MITERS about how you would go about making an octorotor quickly without having to build a full custom controller, and I’m glad to report that it turned out to be awesome – more on that shortly.

But Tinystar won’t be hauling any FS-100s or RED EPICs. Why?

because it’s tiny and adorable!!!

(Tinycopter is on the left, by the way).

It’s a little over 13″ across and weighs about 280 grams with battery. Now that I’ve shown a picture of the final product, this is how it went down. It literally happened this past weekend.

Strictly speaking, it took more than just 2 days, but sporadically so. The design work was completed in a day, and I intermittently printed the frame components on the Lab Replicator (I can actually say that now – are we living in the future?) over the past week or so. This frame uses the same tactic that I settled upon for Tinycopter v3 – 3D printed joists and clamping components stuck to carbon fiber tubing. Many multirotor kits are made this same way (though with injection-moldings instead of 3D printed parts of course) and the arrangement is sufficiently versatile and adjustable to be able to mess with the design quickly.

The two big round pieces form the main center “hub” of the frame, and the small H-looking things are motor mounts. The landing legs are made of 2mm laser-cut plywood, and will be painted to look more badass.

The arms are made of DragonPlate’s 1/4″ pultruded CF tubing, which I bought a fair amount of for Deathcopter a long time ago, cut to length using one of the small bandsaws.

Here’s the two multirotor control boaards. The mounting plates have mirror-offset holes which can be arranged in a regular octagon using standoffs. The two flight controllers are mounted right above each other, with the height chosen such that one is slightly above the height center of the props and the other is slightly below, for symmetry.

The controllers do not interface with each other at all – they are totally independent; if I had enough thrust, I could fly it as either quadrotor. The “copter mode” is selected by a few DIP switches and the control gains by trim potentiometer. It’s important that these boards are very simple gyroscope-only controllers. They do not help the frame achieve self-levelling due to the lack of accelerometers, so it’s a very different experience flying. I’m used to Tinycopter and its (vaguely) angle-controlled flight, so I’m still not very proficient at flying Tinystar.

Incidentally, a week after the “double quadrotor” discussion and of course right after I had ordered two of the i86 boards, Hobbyking unveiled the KK2 controller which has 8 outputs and performs self-levelling (though not quite the same as angle-control). And a LCD with menu system. AND a little piezo buzzer.

Hmph. Oh well – these things were already in the mail by then, so let’s just press on.

The motors for this thing are the HXT 5 gram outrunner motors – in fact the same ones that 4PCB uses. Tinycopter uses the slightly larger 10 gram motors, but I decided that they were too overkill for this project and could not use the 4 x 2.5″ propellers effectively. The 4 x 2.5 props are even a little small for the 5 gram motors in this application, and in retrospect I should have made the arms slightly longer to use 5 x 3″ props (the same type that Tinycopter uses) for more thrust and payload.

#2-56 screws are used to clamp the blocks to the tubing. I want to go back now and replace every one of these with real steel screws. The few grams I gain in weight will be more than made up by increased stiffness, because these screws can’t apply enough clamping pressure before they strip. Tinystar’s arms need a little adjustment every time I pull a rough landing (which is like, every time).

The frame mid-construction on Saturday…

I assembled the frame as the landing legs’ paint dried. The effect is definitely not carbon fibery, but it’s not bare wood. I am a fan of this style of landing gear and may make some (not so high-heeled ones) for Tinycopter. It’ll definitely fare better than my little 3d printed landing claws on it right now…

Starting to look like something…

The weird arrangement of legs is due to the real Cinestar 8’s need to carry a large camera on a gymbal underneath, with the rearward bias of the landing gear keeping them out of the camera frame further. It also looks more badass. That was all for Saturday – Sunday was reserved for wiring everything up!

oh god

Clearly as the number of rotors increases, wiring messiness increases nonlinearly. This is going to be difficult to arrange elegantly and in a vaguely easy to service fashion…

I guess it didn’t end up that bad looking. I pretty much daisychained each ESC together, so it’s actually quite suboptimal from a power distribution perspective – the last ESC in the chain gets the most screwed in terms of voltage drops and resistance. I would also mount these on the bottom side of the frame if I did it again, for easier access. A tree-style (8-to-4-to-2, for instance) distribution method would have been much better, though a bit more weighty.

Tinystar has noticeable trouble starting all 8 motors at once, something which I attribute largely to the daisychain wiring. Several motors pulsing at once mean the voltage on the positive connection can sag alot while the ground (0v) rises erratically, throwing off the ESCs’ starting routines.

The flight controllers have their own ‘decks’, onto which they are secured with fluffy double-sided tape. This is dramatically less isolation than Tinycopter’s block of memory foam, but it seems to work very well for one reason or another. Perhaps accelerometers really are more trouble than they’re worth…

Powered on! It took a few coin flips to get the motors spinning the correct direction. The controller assumes you have certain motors spinning one way because differential thrust is needed to turn (yaw) and if the props are wrongly-handed the direction of spin for a certain thrust differential will be opposite, leading to hilarity.

I tested one “subrotor” at a time – first, the four 45-degree motors in the “X” configuration, then the 90 degree motors. This was just to make sure I didn’t put everything together only to have something die or not work – I had to go back in after the “X” stage and replace one of the ESCs (Side note: the 6A ESCs come in a old and a new version, the latter of which does not work with 3S lithium batteries but the new one does)

Pretty-shot!

It took me a while to get used to flying in rate mode….

…okay, it actually took me a while to dare fly it at all since it’s too pretty and my past first-flights have all ended in sadness. I still have issues with “station keeping”, or holding it steady in one place, but I also have that issue with Tinycopter and more flight hours can only help it. Rate mode is literally 1 integral away from angle mode, so the joystick movements required are much different. One tactic that has come in handy is briefly flicking the stick in the direction I want to move in- this step-changes the attitude since you command a strong rotation and then none. It’s not the smoothest of movements but it gets the job done.

One thing we noticed immediately is that this thing is so stable. I’m thinking it’s due to two main reasons. First, it has more ‘unit vectors’, so to speak. More directions it can move in while only changing one motor’s speed. A quadrotor moving in a direction that is not parallel to its arms needs to spin up or down at least 2 motors, whereas an octorotor can additionally move in the 45 degree directions without having to do the same. Second, there’s six gyros in total because of the 2 flight controllers, so not only do you get parallel readings, but the flight dynamics make for some mean ‘mechanical averaging’ too – the motors and propellers are on the end of long arms which have some springiness, so noisy command impulses can be absorbed. The propellers themselves are viscous couplings to the air (Thrust is a function of velocity), so it will also tend to damp erratic differnces between the motors.

Basically it was found out that I could set Tinystar flying very level and slow, and actually “dribble” it up and down, pushing down on the battery. It sinks a little, but then recovers.

Then I try flying it again and all hell breaks loose. Hmph.

Anyways, because Tinystar was made in mimicry of something used in cinematography, the test videos are a little more artsy than usual. There are, somehow, already two test videos. The first one was filmed indoors the day of:

And the second one filmed outdoors on the Third Ever MITERS Flight Day with even more arts:


I really should think about making a tiny 3-axis gymbal for this thing… once I switch over to 5 x 3 props so it can lift more than itself.

Return of the Kitmotter

chorl1 Comment

Kitmotter, the concept that never quite was.

Originally, I built Kitmotter 0001 to be used on a little display stand in order to show the concept of laminated (coarsely-layered, anyway) rotors for potential hub motor or custom BLDC motor applications. The concept is a direct knock of the laminated hub motors of B.W.D Scooter which was the first to explore the idea of using waterjet-cut steel rings with magnet “indents” allowing for easy placement and waterjet-cut plastic (or metal) endcaps – basically allowing a brushless motor to be constructed without any intensive machining work. At least, it gets rid of the need for a large machine to turn the steel rotors that my designs all feature – I can make a whole post about all the random workarounds that we’ve tossed around for building whole motors. The same idea has been used on a few vehicles besides B.W.D – the Pneu Scooter and its close design relative picofahrrad, and a non-motor application, just to name some examples.

Kitmotter-on-a-stick lasted for several demo/exhibition type events until I took it to Singapore…


Poor Kitmotter.

The two little necks of 1/4″ acrylic just did not like a kilogram of motor hanging off it as it was bounced around my luggage. And with that, the base was retired and Kitmotter was relegated to a fairly simple life of “static item I would occasionally grab of the table to show people”.

That is until I dropped it one day.

With its main source of attention payments (from curious freshmen and ambitious motor builders) gone, Kitmotter was forced to live in the slums of my multistorey handcart of parts and stuff for many months (picture from before it suffered its unfortunate luggage incident)


Kitmotter could never afford to live in one of those new parts drawers.

As my cleanups and project purges were happening, so the slums were being cleared, demolished to make way for new expensive high-rise plastic sorty-bin developments. The first tenants of the new development were the fasteners, who pretty much lived on a level above the rest of the parts beforehand anyway (literally). Kitmotter was temporarily forced to stay with friends on another shelf in MITERS. The future seemed bleak for Kitmotter – forgotten, broken, and tossed away, an embodiment of ideas whose time had come to pass.

Until today, when I found 3 sheets of almost pristine 6mm acrylic in the laser cutter scraps pile at the Media Lab. Almost, meaning some UROP most likely took 1 part out of one of the corners then left the rest of the material hidden in the pile hoping nobody would find it because he’s too damned lazy to drag it back upstairs. I swear I haven’t done this before during my undergrad adventures, people.

Well, no name means no claim, so I quickly whipped up a new display case design while a job for the lab was processing:


Yes, white on white. I know.

This box gets rid of the mounting ‘stick’ as well as removing a very obvious pinch point in the original design. Instead the motor is first mounted to a reinforcing cross, then screwed to the case. There’s more acrylic to crack… not that it won’t be any harder, since acrylic. The case is also shorter than the first

Here’s the parts of the new box after cutting. I picked some scrap dark green acrylic to make the spacer rings from this time. The “KITMOTTER” is actually vector-etched into the top plate on a setting just fast enough to break the white plastic coating. Normally all of this is peeled off, but the separated letters remained white, making for a good contrast.

But before any exterior remodeling, I first went in and fixed one thing about Kitmotter that has been wrong since the beginning: THE SENSORS ARE IN THE WRONG SLOTS. Actually, even worse – they have been in the wrong slots for every motor I’ve built which has internal sensors.

I have previously put the Hall sensors into the slot between two teeth of the same phase i e.g. between A and a, or b and B. The rationale being that when a magnet transition (edge between N and S poles) happens, that set of teeth will “pull” the magnets above it into direct alignment.

I’m fairly certain this belief was just carried over from 2007-2008 when i was first learning How Moter, and then never validated or refuted. It took several very brain-twisting discussions and whiteboard sessions before I finally saw the fundamental error in judgement. I still can’t quite explain it in diagrams or short technical sentences – this post by Amy might be the closest thing. Bottom line is, the concept is true (magnets being pulled into alignment) but the magnitude of the movement required is 60 electrical degrees, and it is only possible if the Hall sensors are placed between two teeth of DIFFERENT phases (different “letters” in the conventional notation). It is still possible to find a ‘combination’ of sensor and phase connections which resulted in cyclic commutation, but the timing would always be too far advanced by 30 degrees (or too far retarded). Either way, not good, and it explained why 1. Kitmotter always sounded like a moped engine, and 2. why Tinytroller had issues with running it because it had no ability to compensate for sensor timing.

The fix involved just shifting the sensors over one slot. Because Kitmotter was never built to be actually used in a vehicle, I just heat-gunned the hot-glue-retained sensors and squished them down over another slot. I placed them between the ab, BC, and Ca teeth this time, like I was supposed to.

With its new green trim rings (for green energy and transportation!!!)*, I closed Kitmotter back up again, with fresh and unbent bolts too.

*shoots self

The new case completed. The large hole on the top plate is designed to clear the motor wires, and the motor itself is mounted only to the cross.

Replacing the motor controller was a straightforward deal since the wiring remained the same – another example of my project “case mods” this summer, I suppose. The controller is a sensored Jasontroller – this one I actually bought from Jason himself in Singapore last January. These are, unlike the eBay controllers, sensored-only.

I can already tell that the sensors are actually correct this time – Kitmotter no longer runs shittily in both directions! It’s much smoother, and the current draw is lower at 36 volts. Previously I was getting 4-5 amps of current draw, which I attributed to the bearings (huge and greased) dragging it down…all, you know, 160W of it. The no-load current has now decreased to only 1.3 amps at 36 volts, which makes way more sense. The minimum loaded motor speed before it starts ‘bouncing’ due to the timing error is also eliminated – switching too soon would cause the motor to jump back and forth as it doesn’t quite have enough torque to overcome the load before the phases switch again.

With this new discovery, I now trust sensored commutation a little more again. And all this time I thought it was just sensored being terrible.

Here’s a short video of the new home of Kitmotter 0001!

But wait, that’s not all.

son of kitmotter

Everyone wants their offspring to have a better life than they do, and Kitmotter is no different.

To fit a hub motor in a wheel, the wheel must not have a center. One of the issues that caused the project to stall out initially was the lack of a “ring tire” or definitive way of turning a stock wheel into one. We bounced all kinds of ideas around, such as wrapping urethane tread around the steel can like B.W.D (later attempted by Jedboard), but decided it was very troublesome and difficult to reproduce. The compromise idea seems to be Pneu Scooter’s “sidemotor” arrangement where the hub motor isn’t concentric exactly with the wheel, but offset from it in order to use its existing bearings.

It’s difficult to explain how some times solutions to problems seem to pop up with no warning in your head. I had originally thought of using a hole saw to clear out the center of a wheel a long time ago, but I quickly scrubbed the idea because how the hell are you going to keep it centered without a mill?

What I had forgotten back then was that hole saws generally have pilot drills in the center. I’m used to seeing hole saws being used in strange milling machine fixtures to make “fishmouth” joints for future welded tube frames. In this case, the pilot drill is not used.

It was during a conversation with Jamison about his latest sidemotor build that the thought of using a reduced-shank pilot to use the existing wheel bearings of a caster wheel as the centering mechanism suddenly dawned upon me. Now that I have hindsight, duh, it was so obvious. It doesn’t even need to be a drill bit – there’s nothing to drill. It literally can be a pin that is 1/4″ in diameter on one end, since most hole saw arbors take 1/4″ pilot drills.

Buying a 1/4″ reduced-shank drill on Mcmaster-Carr, though, was the quickest solution.

So here’s everything. Most of the skate and scooter wheels that I deal with have an 8mm bearing. 5/16″ is literally a hair smaller than 8mm – 0.3125 vs. 0.3149 (for normal hairs measuring about 0.0025″ diameter). It would be terrible if it were the other way around. You can reasonably use a 5/16″ rod as an 8mm axle – which is exactly the intention here. The drill bit would not have a very exciting existence during this operation, since it would just be spinning inside a bearing.

The inside “corable” diameter of the 125mm skate wheels I use (“YAK” type 12-spoke wheels) is about 3.25″ or 82.5mm. This is conveniently a size for which they make hole saws.

I replaced the pilot drill with the 5/16″ reduce-to-1/4″-shank drill bit. The alignment seems to be spot on here. Because the pilot drill needs to go past the wheel’s bottom, I elevated the whole thing on a piece of scrap wood. I clamped the edges of the wheel through to the workbench to stabilize it – a similar procedure will probably need to be done for drill press jobs. I’m actually not sure why I went for the hand drill this time – probably due to torque concerns from the ~1HP DeWalt drill versus a wimpy 1/3HP drill press motor, but the alignment and precise feed control could have made alot of difference.

Alot of positive feedback induced jamming later (note to self: drill press.), and  it worked!

After vacuuming out the swarf, the result is quite splendid indeed. The interior finish is clearly not as refined and clean-shaven as a lathe boring job, but who cares?

Basically, my assessment of this process is that one of the last missing pieces of a fully-accessible Kitmotter has been realized. I’m now really kicking myself for not having thought of this earlier. The major problems with Kitmotter were the lack of consistent stators (an issue that is solved by consistent copiers and laser printers), stator-to-bore adapter solutions (solved by 3d printable nylon hubs you can get from Shapeways or other 3DRP vendors), and… tires.

The same type of 12-spoke YAK wheels also comes in 100/110mm size which I’ve confirmed to be borable out to about 2.375 (2 3/8″, or 60mm). Guess what – they make hole saws for that size too. This can get exciting.

The hole saws actually appear to cut a little oversize. There’s two contributions to this (rather massive) overcut. First, the saws themselves are a little bigger than nominal dimension, and second, I could not hold this thing straight at all manually. It seems like a drill press is pretty much mandatory. Each time I twisted the saw, it would take a bigger bite out of one side. It tended to jam on the thin plastic spokes (a finer tooth saw would also mitigate this), so some times the twisting was fairly severe.

However, taking into account a straight cut, the diameter is probably going to be still 3.26″ to 3.27″ anyway – which is pretty much 83mm even. Sloppily using an Inch tool to make a metric dimension….

So, what’s the next step? I need to reduce Kitmotter 0001’s 3.5″ outer diameter down another 1/4″. This is much more difficult than it sounds, because the magnets need to come down in thickness (probably to 1/8″) and the can must get substantially thinner radially, and I could run into trouble with the case fastening screws. The screws will most likely need to be moved ‘external’ to the steel ring, sitting in circular grooves.

Alright, enough of this “future talk”, here is Kitmotter 0002, coming soon to a…nother demo stand? near you.

So what’s going on here?

  • The 7 #4 bolts have been turned into 14 #2 threaded studs. This diameter change was absolutely necessary to thin the can down down to 3.25″ – there was no other way.
  • The radial grooves seat the screws and secure the layered can, while the endcaps have fully enclosed holes to keep the studs on-dimension
  • The magnet thickness is 1/8″ instead of 1/4″, again to bring down the diameter. I’ll definitely lose a little torque from this.
  • The bigger endcap has a flange that is supposed to be used as a guide to drill into the wheel, in order to retain it. I might add more holes for more strength since the threads will be in soft gooey scooter plastic.
  • The axle is a stock 5/8″ keyed shaft of a 3″ stub length that McMaster sells directly. The hub to the stator is 3d printed. All other trimmings are left up to the user.

I’m going to build a prototype of Kitmotter 0002 in the next 2 weeks to validate this model, and I consider it right now to be prime Instructables fodder if it ends up working out. Essentially the last missing link in an “accessible” hub motor vehicle has been solved – where accessible means hypothetically buildable without access to heavy machinery like lathes and mills.