Saturday, May 26, 2012

FFv1.1: Air, Land, and...well that's it.

FFv1.1 is the new minor revision of my small motor controller, which has so far tested successfully on the Cinestar hexrotor motors. The controller is sized for 24V (6S LiPo), 30A continuous and 60A peak current, and uses the same Sensorless, Sine-Commutated Field-Oriented Current Control (SSCFOCC) that I've been using on Pneu Scooter.

Unlike most of my other controllers, the FF series has power and logic integrated on a single board, optimized for size. This is enabled by the TI DRV8301 pre-driver, which so far has been working very well despite my fear of integrated power/signal ICs. Since I haven't had any obvious hardware problems, the only change in the v1.1 minor revision is the layout of the DC bus capacitors:

Instead of two 10mm electrolytic capacitors sticking up at a weird angle, I opted for a bank of large 50V ceramic capacitors, eight in total, on the board. The ceramics are more expenisve, but they should also improve the high frequency filtering of the DC bus. Because they are much higher cost per energy, I also planned to use a single large electrolytic capacitor sticking off the side of the board.

One problem with developing a controller that might be used on a hexrotor is that you need six of them. So, for v1.1, I got a 5x2 panel of boards from MyRO, a not-shady Chinese board manufacturer (~2 week turn). At this point I should thank the folks from FreeFly Cinema who donated the Cinestar frame and backed the first round of FF controller development.

Here's the fresh panel:

The individual boards are v-scored, so they're easy to break apart by hand. One reason I wanted to get the boards as a panel is so that I could batch-process the soldering. And damn, it was a lot of soldering.

I mostly solder everything by hand with a magical Weller WD1001 station, but in order to do the DRV8301's power pads, I used paste and a reflow oven.

95% finished Side 1. This side has the XBee radio headers.
95% finished Side 2, with the microcontroller and gate driver.
In total, it took about 14 hours of soldering spread over three days and not including the time to attach wires and connectors to the finished boards. It's probably the second longest soldering job I've ever done. But I think I probably cut the time in half by doing the whole panel at once instead of working with individual boards.

The finished product, without heat shrink covering and connectors, looks like this:

The three blue wires are motor phase outputs, the red and black are DC inputs, and the three-wire connector is the servo-style PWM input. There's also a 6-pin header for programming with an FTDI adapter, although so far I've been programming wirelessly by XBee. In this configuration, the total footprint not including wires is 4.25"x1.5"x0.5". Without the programming header and electrolytic capacitor, the minimum footprint is 3.0"x1.5"x0.5". The weight including wires and connectors is about 65g.

For the six boards that will wind up on the hexrotor, I used IRFS3004-7 FETs (40V, 1.25mΩ) and a 680μF/35V electrolytic capacitor. The remaining four use Infineon FETs, IPB034N06N3 G (60V, 3.4mΩ) and 330μF/63V capacitors. These are prototypes for a higher voltage, lower current version, and in particular I wanted to test them on some ground-based vehicles. I left these four connected from the panel:

Quad-core motor controller.
The reason for this is because I wanted to test them on a 4WD vehicle:

The vehicle, ChibiKart, is a 4WD miniature go-kart with custom in-wheel motors. The controllers are shady eBay specials that do sensorless BLDC at about 25A peak for $28. While I definitely can't compete with them in cost, I can match their power and four FFv1.1 boards could fit in a single one of their enclosures. (I may actually try this, now that I think about it.)

It's probably a better idea than a McMaster-Carr bag.
After some initial testing (video below), the controllers performed well with 25A peak current but had trouble ramp-starting all four motors at the same time. The eBay controllers can do it on level ground, but with limited torque for the first instant while the back EMF zero crossings are still noisy. My eventual goal is to have a sensorless start at full rated torque, although it may require more than simple open-loop ramping. For scooter (or e-bike) use, the simple ramp start or a manual push start would be sufficient.

Like most of my motor controllers, FFv1.1 is set up for wireless data acquisition. Here's some data from one of ChibiKart's test runs:

The regenerative braking on pedal lift had a nice feel. I'm eager to implement a better ramp-start and get tinyKart running on sensorless DirectDrive like I've been saying I would do for months.

But for now, back to multirotors. The main purpose for the other six FFv1.1 boards is to test fly on the Cinestar 6 hexrotor. Before I feel comfortable enough to do that, though, I decided I should flight test the controllers on my Turnigy Talon quadrotor. It's a much more manageable size (500mm from center-to-center), but it can carry a substantial payload so it wouldn't mind the extra weight of large controllers and heavy-gauge wiring.

I was going to heat shrink the controllers, but decided to leave them bare for testing until I'm confident they don't have any bad solder joints. So, I just zip-tied them to the frame for now:

To accommodate the output signal from the KK multirotor control board, I wrote a simple PWM input routine using a spare timer on the STM32. It measures pulse length and adjusts voltage magnitude directly (no current control.) However, the voltage phase is still feedback-controlled to keep d-axis current at zero. This is a unique control scheme not possible with normal Field-Oriented Control. It should retain the "feel" of an airplane-style RC ESC but have optimal torque per amp at all speeds.

For the PWM input, I used a digital low-pass filter to limit the spin-up rate of the motors. I initially did this in lieu of current control to limit the maximum amount of current the motors would see under normal operation. However, after flying with a relatively fast input filter (τ = 33ms), I encountered the other reason why one might want to filter the input: vibration noise.

The PWM input under hover conditions would actually fluctuate by about +/-50μs (+/-8% of full range). This noise is almost certainly from the gyros on the KK board, and I don't blame the cheap controller because I had major gyro noise issues with the Pololu minIMU-9 too on 4pcb. Gyro noise due to mechanical vibration has been by far the most hassling part of building multirotors for me. In any case, it was actually beneficial to increase the input filter time constant to about 50ms and turn down the control gains slightly to reduce how much the ESCs would amplify the gyro noise.

Lastly, video of both ground and air testing:

So far the controllers are working as planned. This marks the first time I've ever tested eight controllers in one day without blowing anything up.

Sunday, May 13, 2012

2.007 = Wide-open spaces for tinyKart and Kramnikopter!

Oh yeah, there's also the whole robot contest thing:

It's like mini-FIRST.
For the past four years, I've been a teaching assistant for 2.007: Design and Manufacturing I, a sophomore-year mechanical engineering design class where students each create a small (16"x16"x12") robot to compete in an annual competition. This class, formerly numbered 2.70, is the origin of the FIRST Robotics high school level competition. If you don't believe me, here's Woodie Flowers with a giant camera octopus:

Just like in FIRST, there's a new challenge every year, and this year's game was Tech County Fair. Robots scored points for collecting arcade tickets, filling a balloon, hitting a high-striker, and spinning a scale Ferris wheel. During the competition, my main job is running the scoring computers. Here's a view from my desk down on field level:

This year, only about 76 robots (out of a class of ~140) entered the final contest, which is optional and doesn't count toward their grade. On one hand, I would like to have seen more of the robots finished. But, having fewer robots in the event means that the ones that do compete tend to be very good and the matches are more exciting. Here's a picture of the winning robot pair:

The bot with hammers on top of it would hit the high-striker (not with the hammers, those were just ballast). The smaller crab-looking bot would pinch the Ferris wheel and spin it, as shown in the picture. Both robots could autonomously find their targets and start their tasks, a huge advantage in the final rounds.

Two other really cool entries that unfortunately did not make it to the final rounds were the Whitelaw Prize (design award) winners:

The bot on the left had several unique features that I've never seen on 2.007 robots before. Instead of striking the lever, it would lift the high-striker mass with a tape measure-like constant force spring that extends from a roll inside the robot chassis. The green Banebots wheels look cool and provide tons of traction. (Are they even kit-legal?) But the real surprise was that the wheels could be steered much the same way as Twitch, having a 0º, 45º, and 90º orientation. The steering was accomplished with crossed-over belts instead of linkages. Also, the whole thing was controlled using DTFM tones from a telephone, just because.

The bot on the right has a more conventional but very well-built 2WD drivetrain. Its showpiece is a custom-made steel flywheel that spins up to a few hundred RPM. It released its stored energy into the high-striker lever so effectively that this robot damaged pretty much all of the practice levers and they all had to be replaced for the contest. It could also spin the Ferris wheel using rubber-coated rollers on the back.

You can watch the two-night competition at the following links:

At the intermission of the Finals (1:55:20 in the second video), you'll get to see the other reason why I enjoy the 2.007 final contest: three days of access to a huge venue with high ceilings. I brought 4pcb and Kramnikopter to the event and while I was testing out the $35 HD Wing Cam I just got for Kramnikopter, the AV crew asked if it could lift their GoPro. It turns out that the Talon frame is pretty much a perfect match for the GoPro, and I got to do some flyover videos of the contest field. Here's some comparison video between the cheap but lower-quality HD Wing Cam and the GoPro:

The HD Wing Cam shoots in 720p and really needs good lighting to work well. But it's small and cheap. The GoPro shoots full 1080p and has much better color balance and image stabilization, but it's $300. It really seems like the way to go for this size quad. They didn't let me keep it, though, so a few days later I went out with my HD Wing Cam and did some more outdoor testing:

You can catch a few glimpses of the giant hexrotor, which flew for a bit too!

Another new 2.007 event this year was an official Electric Vehicle Section final contest. We've had an EV Section (optional, for people who are bored with robots and like EVs) for a few years now, but there hasn't been a separate final event specifically for the vehicles. This year we ran a 50-meter drag race and a unique urban hill climb in a parking garage. (Both time trials, no actual racing for safety reasons.) You can see a full summary of the event on Charles' site. I'll just include the video highlights:

tinyKart, Chibikart, and Pneu Scooter also did some timed runs, since the instructors should be allowed to have some fun also. tinyKart did a 6.52s run on the 50m drag race, about what I would have predicted. It also did a 53.9s run up the garage with a peak power draw of 4.1kW from the batteries. I intend to finally convert tinyKart to running my DirectDrive controllers, so it's possible it will get a performance upgrade soon. (Or there will be a lot of DirectFET smoke.)

Coming soon: FFv1.1, dynamometer, Gen. 1 sensorless write-up.

Saturday, May 5, 2012

KKv0.0 Camera Platform

After a not-so-successful attempt to mount my HD video camera to a Turnigy Talon quadrotor frame, I concluded that the camera is a bit too heavy for this configuration, and that better vibration isolation material would be needed. A GoPro is much more well-suited to the task, so maybe that will happen in a future version. But I also ordered one of these $37 720p minicams, which have gotten mixed reviews. It's much lighter than either of the alternatives and I would be much less sad if I lost it in a crash.

For now, though, I decided to explore vibration isolation options without solving the problem of having a too-heavy camera on this motor/prop combo. The two materials I tried in v-1.0, felt and closed-cell foam tape, both failed to damp whatever frequency of mechanical vibration was causing the camera to be shaky. So, I went for a known solution of memory foam (e.g. McMaster PN 86195K314).

There's a 1/4-20 screw going into the bottom of the camera that, with a large washer, sandwiches the camera mount in between two pieces of memory foam. Here's what it looks like on the frame:

The improvement in video clarity was impressive, though it didn't eliminate the vibration completely. The camera is still too heavy, so the whole thing is hard to fly and the motors and ESCs get hot. But here's the result:

Thursday, May 3, 2012

Cambridge Mini Maker Faire + Kramnikopter

A couple weeks ago I went to the Cambridge Mini Maker Faire. It's a smaller event than the full-scale NY Maker Faire I went to in September, but it's much closer to home. In fact, it's within walking distance, so I decided to kart everything over.

Yes, kart.
It turns out that tinyKart makes a pretty good hand truck. Also along for the ride was Pneu Scooter (ziptied to the frame), Twitch, and 4pcb (in the seat). Unfortunately, the Mini Maker Faire is too crowded for the vehicles to be effectively (or safely) demo'ed. And it was too windy to fly 4pcb.

So that just left one option:

Balloon Twitch.
You can see how windy it was by how far the balloon is deflected. The balloon was my way of making sure that nobody stepped on Twitch accidentally, but it also made the robot a lot more "interactive". And by that I mean that several little kids were teased.

Sorry for the blue video...forgot to turn off manual white balance...

You can find more photos from CMMF 2012 in the Flickr pool.

I was really looking forward to demo'ing 4pcb, since it drew so much attention at the NY Maker Faire despite not actually being flyable at that point. It's small enough not to pose much of a safety risk, but micro quads really don't like wind. In general they're twitchy and hard to fly. Giant hexrotors have the opposite problem: they're very stable, but too dangerous to use anywhere near people.

To fill the gap in my flying things fleet, I decided to put together a Turnigy Talon frame from Hobby King. I got to fly one that was put together by Daniel Kramnik (of Tesla Coil fame) and was very impressed with the quality of the frame. For $34, you get a real carbon fiber + aluminum frame. Add to that four $7 motors, four $14 ESCs, and $18 battery, a $3 bag of props, and $15 control board, and you get a complete mid-size quadrotor kit (bring your own radio) for $154.


Here it is with 4pcb propped on it for size comparison:

The Talon frame is very impressive, even without considering the cost:performance ratio. It's stiff and light and it looks amazing. It also seems very durable - you'd have to crash pretty hard to break it and the landing gear is nice. When you take into account the fact that it's $34, it is one of the best deals I've seen.

The motors, on the other hand, leave something to be desired. They are the most inexpensive of an already low-cost/low-quality brand name, which means they have some issues. Mostly, I ran into axial alignment problems - the can and shaft are poorly constrained. They are also not balanced. If I were to replace one component right now, it would be the motors. They do look cool, though. 

For what it is, the KK board is an impressive deal as well. It's rate-mode only, so it can't do self-leveling, position hold, altitude hold, or any other more advanced features. 4pcb flies in self-leveling attitude mode, so returning the sticks to center means it tries to go to level (zero angle). So it took some practice for me to learn how to fly Kramnikopter in rate mode, where returning the stick to zero means zero angular velocity. You can see me learning the new input mode a little at the start of this video:

After filming the first half of that video with it, I decided to go ahead and attempt to fly my HD video camera. I have a long history of attaching my camera to things that move, but this would mark the first time that it's actually left the ground. The mount I made was pretty simple:

Kramnikam v-1.0
On one hand, the camera flew and survived. But it's not quite a cinematic experience yet. This camera weights exactly 300g, and it's just a little bit too heavy for this combination of motors and props. The resulting flight is very close to being unstable, since the motors are almost maxed out. The more obvious problem is that my initial vibration isolation solutions (foam tape, felt) didn't help much. So, lighter camera and softer mount seem to be the next steps. For now, I have a really nice medium-sized quad to play with.