Interest in an universal Dynamixel device?

[xevel]

Hi!

For a new project of mine, I finally decided to work on what will be, in effect, a universal Dynamixel device board.
What I'm looking for here is some feedback, to see if people have a need for it (it will be completely open source but might not be the easiest to assemble...), and therefore if there would be things I could change to make it even better for the community.

For too long, when we wanted to put a sensor or a micro-servo at the tip of an arm made with Dynamixel servos, we had to put a lot more cables, and bring everything back to a central controller (that in turn needed GPIOs/ADCs...)


The plan:
- really Arduino-compatible, with a real ATmega328p so that nearly every lib written for the Arduino could be used as-is (NOT a half-baked port to a new arch).
- as small as possible so that it could be put in robots, even at the end of limb
- powered from the Dynamixel bus, and capable of providing power efficiently to the devices you want to hook up to it (around 500mA)
- support for the complete Dynamixel protocol @ 3Mbps, without putting any unnecessary burden on the Arduino-compatible part.
- 5V
- a few pins with 0.1" 3-pin headers GND-5V-SIG so that sensors and little servos could be plugged in directly.
- fully open : https://github.com/Xevel/dynext

Some of the cool part I would love to have, but may be discussed or may take some time to develop
- precise voltage ref for the ADCs
- programable from the Dynamixel line
- multiple virtual devices: hook up multiple components to the board, and virtually give each of them a different ID so that they can be adressed by the master more easily.

Arduino compatibility seems to be maybe the most important part. I want this to be as easy to use as possible, so no need to install a new toolchain, no new architecture to learn, as little porting of libraries as possible (ideally, none, but I won't have enough room to put all the regular Arduino pins...)
I'm interested in feedback on any of these points

Other boards with similar descriptions have been made, by HUV Robotics, iBot has made some I think too, and others (I have made some for my Xachikoma bot a few years ago), but most I've looked at used a lot of processing power to listen to the bus, and none could work at 3Mbps that I know of. Plus they either were not easy to program for non-gurus, or simply not reprogramable.


Anyway, here is what's going to happen:
I need these for my next bot. I'll put at least 6 or 7 of these, one per leg (it's a quad), plus one that will control a motor and its sensor, one that will ouput some servo PWMs, and one that will control a camera subsystem.
The goal is to cut on the wiring, have a better layout of the bot (control where it is needed instead of everything in the same place), put some local Dynamixel loops in each legs (AX-12 with REAL torque control )...

If in the meantime I can make it better for other people's requirements, it's a win-win!
Along the development, some beta boards will be available around "at cost" for people who might be interested.
If there is some interest, I'm tempted to make a small crowdfunding campain just for the heck of it ^^

Where is the project now:
I have a first version assembled. This one is a first shot, it is not Arduino compatible, and doesn't really fulfill most of the requirements... but it allowed me to test a few things, and to show it around to some people at Innorobo last week - who are now very interested for their own robots, some even commercially.

The first version (Dynext v1.0) was ATXmega based, used a one-chip solution, was 3.3V and not 5V tolerant, and had ADC with a strange range. The layout of the pins is also not that great when thinking of plugging a servo or a sensor: it nearly needs a secondary board to be made with the right connectors and wiring for power. This one can be found in the Git repo linked above.

Dynext v1.0

dynext_v1.0_proto1.jpg (190.38 KiB) Viewed 39213 times

The current version looks a little bit like that (component selection not finished)

Dynext v2.5 preview

dynext_v2.5_prev01.jpg (125.42 KiB) Viewed 39213 times

It's a 2 chips solution, like the Arduino Uno: one is dedicated to talking fast (a very small and cheap Cortex-M0 at 50MHz), and the other one (atmega 328p) is for the user application.
The size of the board should be around that of a Teensy 3 / Arduino Pro Mini (same width, a little longer). The two right-angle dynamixel connectors are underside so that when the cable is plugged it does not change the actual footprint nor the whole thickness.
To fit everything I'm going to use very small parts, but I'm convinced it's worth it : putting half a dozen of these in my very small new bot (the body will be around 12cm accross!), every millimeter counts.
People willing to replicate can use bigger packages if they want, every component has a hand-solder-friendly version available.

So here it is. As soon as I get something I'm happy with I'll order new PCBs.
In the meantime I'll develop the software with an LPCxpreso board and and arduino.

Again, I'm VERY interested by feedback.
Cheers

[siempre.aprendiendo]

Only a few times surveys in this forum get a significative amount of answers, so take it easy.

I think it's a good idea and I would buy it. In which price range are you thinking?

By the way, I think that could be very useful for the new Darwin mini!, that should be launched in a few months.

I'm learning 3D design, so if you add some more data, like height, holes distances and diameters I would design , print and upload a little case design for it.

[xevel]

Hey, thanks for the feedback. I don't really expect loads of people answering, I have seen how little action the forum has these days. I just hope some of the regular gurus (like you :p) will see it and pitch in.

In terms of price range, it's still something I'm not sure about. I think having something that actually works very well is more important than saving 3€ on the final price... so I'm considering using some nice parts to have good efficiency, good ADC stability, and enough power to put two mini or micro servos, at least. It's a real challenge to get the power aspect right on such a tiny board.
I'm my first consumer, and the next ones are friends who actually build robots they intend to actually _use_ for a greater purpose, not only play with (think researchers).

Making version with the XL-line connector is something I would like to do too. I should get mine soon, and plan to use them in the same bot as this. The circuit and components could stay exactly the same, no pb.

Currently, I'm working on a proof of concept with the new architecture, and I have to test some new (very small) components, to find a good balance between size, cost, efficiency, etc.
I'm looking into providing maybe half an amp of "dirty" 5V for powering servos, some "clean" 5V for the ADC part, and some 3.3V for the Cortex M0. I'm probably going to wander in the "designed for cellphones" aisles of Digikey/Mouser/Farnell for that, it's going to be fun :p

[ttuuz]

Hi cool looking project haveing just built a atmega8 16pu versoin of a dynamixal device my self i am very intrested in the one your building nice looking job so far.

Cheers Ray.

[PaulL]

I'm looking into providing maybe half an amp of "dirty" 5V for powering servos, some "clean" 5V for the ADC part, and some 3.3V for the Cortex M0.

Half an amp isn't much, do you intend for these to be micro / nano servos? The Hitec HSR-5498SG's I am using are some 300mA at idle (x20 in my case, so 6A to idle all servos), I forget stall values - 2 or 3 A or something. Not sure what the MKS micro servos I'm using are rated for - I chose a 12A DC to DC converter for 13 of them, which is probably more than I need (though, the max amp rating will likely de-rate a bit from less-than-optimal air flow).

Nice to see others doing some board work - I think that's one aspect that's been missing from the hobby-scale bots. And, it's become much easier and economical to dev a custom board these days.

I plan to do something similar in my bot, comms to STM32 boards in hands / feet - the STM32 has a processor interconnect feature via UART, I'll be fiddling with that to keep the wires down and bus speed up (though, the RM0007 is wrong in one aspect - refers to AND for tx from slave processors, should be OR). I figure there are about a zillion options to having a processor / MPU-9150 in hands and feet, so I'll end up doing some kind of add-on plug for sensors / devices (I2C, a PWM / GPIO pin or two, and power).

Regarding power - I had some early ideas that were a bit less than what turned into a full blown power board, but I kind of went nuts with it. The more I thought about it, the more I realized I wanted more control over power. It doesn't help that there are thousands and thousands of "small" regulators - some useful, some not so much.

Paul

[xevel]

Indeed, 0.5A is not going to power big hobby servos, but that's not the idea. If you can put a big servo, then you could put an AX-12 or have room for a suitable power adapter. The servos you use are nowhere near what I imagine for that.
It's an arduino on the dynamixel bus, not a power board; you wouldn't try to power your big servos directly from an arduino.

The goal is more to be able to put one or two micro servos to move some sensors or cosmetic parts.
Not sure how useful that might be for the general public, but i have had the problem in Xachikoma.

For power, in my bot, I'm making a full blown power board, but it's completely tailored form my bot. I'll release it in open source as a curiosity but I don't expect that to have much value for anyone else.

[PaulL]

Agreed.

For power, in my bot, I'm making a full blown power board, but it's completely tailored form my bot. I'll release it in open source as a curiosity but I don't expect that to have much value for anyone else.

I'm in the same boat. There just aren't any suitable off-the-shelf power solutions out there, and my power board is focused on my particular bot - I doubt anyone would want specifically what I've done so far, either. What IS available usually ends up being too big or the specs aren't enough.

To get more into / out of these bots, the electronics have to be tight, have to be in as small a form as practically possible - and about the best way to do that is full custom designs (though, I do wonder if given a particular open source design if some would bother to route to their required form - maybe a few might do so). Initially, I thought I might get away with piling a few break out boards together, but that quickly became unrealistic, just not enough space. In my power board, the width I needed is very narrow, just enough space to route signals - barely.

On the bigger picture, I still firmly believe in a bot controller (servo control, sensors, power control, etc) as a USB-interfaced solution. A controller for a particular bot can only offer up but so much capability to the hosting processor board, but the options for SBC's are wide and varied. I see a bot as a pile of hardware, with the controller as the interface to that hardware, with USB presently being the most widely available / standardized interface to a host processor. I don't think a single microcontroller running an entire bot is enough anymore. I think of the bot more like a printer, a peripheral with its own internal microprocessor, where the microcontroller tells the hardware exactly what to do based on what it is told from the host system. It's a model that has done well in the past. The difference for me is that the host system is along for the ride, so to speak.

That said, if someone was interested, I'd modify my controller to suit as a stock replacement for the RN-1, but that's an entirely different direction. Maybe the RasPI is small enough to piggyback, but the stock case wouldn't fit.

As for the Roboard RB-100, I am one of only a few people that have put one on a Robonova, maybe the only one to do so under WinXP, and even then, it's limited in one too many ways for me (full duplex audio problem for TTS / VR was what made me abandon the RB-100).

Keep up the good work!!

[xevel]

I think I have found how I'm going to do the power part. At least it's looking good, but I haven't finished reading the fine prints of each datasheet.

- Switching reg @ 5.4V, 1A (TPS62160DSG)
- Dual adjustable LDO: 5V@300mA, 3.3V @ 300mA (ADP223)

5.4V is acceptable to the IOs of the ATmega running at 5V (limit is VCC+0.5V), and all these have small footprints.

Paul: Oh how I understand what you're getting through...

[xevel]

New version of the board done, prototype boards are in production, and I'll order the last missing parts in a few days, they arrive in less than 24hrs anyway (Farnell rocks: order before 7pm and get the parts the next day around noon, free shipping and payment at 30 days when you have a payment account ).

dynext_v2.0_top_view.png (153.69 KiB) Viewed 38922 times

dynext_v2.0_bottom_view.png (122.56 KiB) Viewed 38922 times

dynext_v2.0_schematics.png (40.06 KiB) Viewed 38922 times

PDF of the schematics

Repository with Eagle files: https://github.com/Xevel/dynext

Power:
I finally went with the 1A switch reg previously mentioned (TPS62160DSG), that will be adjusted between 5.2 and 5.5V - I have to test what is the smallest value that gives satisfactory results. For 5V and 3.3V I changed to two separate 200mA LDOs ( TCR2EE50,LM for 5V and LD6805K/33P for 3.3V), it's cheaper, and smaller when you factor in the resistors that the adjustable dual LDO would have required.

Connectors:
The main connectors expose A0-A5, 0-3 and 9-13 of the Arduino-compatible part.
11 and 13 are used for SPI and have a voltage divider to go from the 5V of the ATmega to 2.5V for the LPC (that's enough to register as a HIGH level on the LPC11A02, since its HIGH threshold is 0.7xVCC = 2.31V).
I also moved the LED from 13 to 6 to make it actually available to the user.
5 (called SSEL/ISP/5 on the schematics) is used as Slave Select for the LPC in normal operation, and can also serve to put the LPC in ISP mode if held low when the LPC resets or boots up.

A0-A4 each have SIG-5V-GND connectors, on the "clean" 5V rail (limited to around 150mA).
9 and 10 have SIG-5V_DIRTY-GND connectors, on the ~5.4V directly from the switcher (limited to around 800mA). This makes it possible to power two small servos or power-hungry devices that does not require precisely 5V and can work with a little bit of ripple.

On the top of the board, there are two 4-pins headers with GND-5V-GND-3.3V and GND-RST-AREF-A5.

For the LPC, there is a 1.27mm header to reprogram it with SWD. It does not follow the standard as it would have required a 10 pins connector (too big and half of the pins would have been unused), yet exposes the necessary 3V3, LPC_RST, SWDIO, SWCLK and GND.
There are also test points for the various signals of the LPC, and one (labelled TP) that is only here as a GPIO (useful for non-intrusive debugging and performance evaluation).
Under the board, there is a 1.27 set of pads with SSEL/ISP/5, GND, TX and RX, to make it easy to reprogram the LPC using the serial bootloader that the chip has in ROM.

As you can see the silkscreen is a little clipped here and there (AREF, GND, RST are hard to read), I'll try to fix it in a revision.

I should get the PCBs (1.6mm, 4 layers, gold plated to avoid problems with the BGA) in a little more than a week.

[PaulL]

Looking good!!! The worst part for me was waiting on the boards to arrive, everything else showed up first.

One of the things I've encountered is that it seems you can never have enough power.

Do you plan on using a solder paste stencil, are you going to hand solder these? Just curious. I used stencils from Pololu.com, the price was well worth what likely would have been a lot of aggrivation otherwise. That said, I ended up hand soldering a few parts (miss-order), which weren't too bad (hot air reflow). Rohs, or leaded? I have to say, the leaded stuff I used was really easy to work with (Kester Type R276 CAT#7016070520). If I were to sell, I'd try to go lead-free, but I myself wouldn't hesitate to buy a leaded board. Any plans for how you're going to reflow this thing? Skillet, toaster oven, hot air tool?

FYI, if you add your files via File Manager (off the green button top of page), then do [ img ] and [ /img ] (without spaces) around your file url and path (http://robosavvy.com/Builders/Xevel), it'll populate to the pics on the main forum page. I always preview to make sure I haven't made a typo in the pic URL's. Also, doing it this way doesn't limit you to the scrolling box for the image (but you don't get the image view counter, which doesn't mean a whole lot anyway - bots).

[xevel]

Hey Paul,

Thanks

The boards should get there around Thursday. On cheap oversees pooling services, 4 layers prototypes are much more expensive than 2 layers, which nearly make the all-around more expensive European options less unattractive. I ordered at Eurocircuits this time, which has its advantages (awesome preview tool, fixed delivery schedule, cheap shipping).
I'll order some more parts on Farnell on the day the boards ship [the more I wait, the less things I forget for all my projects!], and I'll get both the next day... UPS power!

I'll put solder paste with a stencil. I get mine from www.smtstencil.co.uk, it's cheaper than Pololu when you factor in shipping. I've tried the ones from Pololu a long time ago, I can't remember the quality of their offering, but the guy from smtstencils recently upgraded his laser cutter to something really nice, changing the previously "OK" quality to "very good". Openings for an XSON-6, with 6 pads (0.35x0.3mm, 0.5mm pitch) on a 1x1.5mm surface, are now very clean (they were previously "good enough" but resembling potatoes more than rectangles, not anymore!).

I use leaded paste (MCN-300, sold for cheap in 50g quantity anywhere). I overstretched the life of the first container I got up to around 2 years (pre-heating the container in hot water before applying make it possible to use old, half-solid paste, which is good enough for big parts). I finally got a new one recently, and this time I'll simply throw away any paste that has left the container and has not been used, to avoid small chunks of solidified paste making the process harder - I can't go through 50g in a year anyway, even throwing 60% away like that.
I tried unleaded, got a rather large quantity of it inadvertently, but I did not push the test far enough to get to a reliable result. I'll let that to assembly houses that do reflow soldering in a controlled environment. This one is quite dead too, and it's fun to plunge your soldering iron directly into, makes a LOT of nasty fumes and cool big solder balls ^^'

The first time I assembled my own small SMD stuff (small = 0402 and QFNs, bigger than that does not need anything else than an iron), I did it with some professional tools (manual PnP, IR reflow oven...). Yet I quickly discovered that tweezers and anything that could heat up to 260°C was enough. Around 40 of my USB2AX prototypes (15x20mm), as well as some other slightly bigger boards (up to 80x50mm) for other people, have been assembled like that, using... a 150W garden halogen lamp :p Works a treat, you just have to move the board constantly on the glass to make it follow a temperature "profile". I roughly mapped the temperatures on the lamp (with crazy gradients!), and for some time this <10€ solution proved to be enough. At the beginning I've had a few boards failing in the field (= dead on arrival when sent to beta testers at the other side of the world) due to cold solder joints, but I pretty quickly got the hang of it. Failure rate after adjusting the process under 10% with 0402s and 0.5mm QFNs, most detected visually and fixed with a touch of the iron (the origin lies in the paste application, I need to make a proper jig for that instead of using tape and old PCBs...).

Nowadays, I invested in a small oven with forced convection, and while I have all the parts needed to make it automatic, I just use it with a thermocouple on a multimeter. Losing 5 min (monitoring the temperature and adjusting the thermostat manually) every 2 weeks is not that bad compared to getting or building a proper PID controller, and the result are quite nice now that I have an idea of the thermal capacity of the oven.
For the USB2AX, now that it's a real product, I have it assembled professionally, it's cheap enough and the quality of the lead-free soldering is great.
For rework, I also have a cheap hot-air rework station (VTSS110, pretty shitty quality but works). Even that one is AWESOME when you have to desolder a 20-pin chip, or a QFN... In the beginning I did some desoldering with a hot-air gun used to remove wallpaper... too hot, too big, but just the thing you want if you're removing indiscriminately all the parts from a CG card quickly :p

Thanks for the image tip ^^ I'll try that sometime.

[PaulL]

Hey Xevel,

Thursday, not bad. OSHPark takes a bit longer (about 10 days last order), especially with economy shipping (I'm in the USA - don't hold it against me!), for a 2 layer board. 4 layer is a good bit longer than that.

I get my parts from Digikey, they tend to show up pretty quick (2 or 3 days) - UPS also.

The quality on the Pololu ones is decent, but not great - good enough for prototypes, though. An MLP28 (CP2103) is pushing it - the apertures look close to the limits of their laser, longer edges aren't very smooth (large surface mount pads for the DC to DC converters).

So far, my parts are hand-placed, but my hands shake (medical thing), so 0805 is tough, and aligning pins for QFN's is even tougher. I cut a venturi vacuum on my lathe a while back (and have a valve and such), I've been thinking maybe I ought to use a webcam and vacuum pickup to place parts jogging my CNC mill (I've seen guys do complete pick 'n place with their mills, I doubt I'll go that far). I've done through-hole and I've built SMD boards (a job I had nearly 20 years ago!), but nothing recent - I had a lot of "what if's", but essentially I followed Sparkfun's method for a reflow skillet (Oster 12"), and things turned out pretty well. I smeared a bit of paste on the corners of an alu sheet (board was kapton-taped to this sheet) and waited for all 4 corners to flow along with the components, then waited a minute or two more for good measure.

150w lamp - reminds me of a thought I had some years ago about a vacuforming machine using 500w halogen utility lights and black sand as a heater. I'm surprised a 150w lamp works!!

Haven't made enough boards to speak of failure rates yet, no product (and I'm not sure that I'd want to sell a physical product, but who knows). I fried a few STM32's on Maple Mini boards due to bad ground on a 12v supply (power for DC to DC Converters), though. I've since replaced them and reprogrammed their bootloaders. Those are Rohs compliant, and takes a LOT of temp to get the chips off (Aoyue 906 Hot Air) compared to leaded solder. Putting them back on took a lot less heat. The controller is good to go, fried one STM32 at first, but it was a breadboard mistake, not the design

Your oven - toaster style oven? I've seen a few guys that have done some nice DIY reflow ovens, I've thought about it a few times. The skillet, though - very easy, heats the board directly - no problems with this approach yet (not even monitoring temp), and you saw the types of devices I had on there!!

When I first got the Aoyue 906, it wouldn't power on. I took off the case screws, put them back on, and for whatever reason, it works. It got a mind of its own once and heated the barrel so much that it glowed red, so it's not that great of a tool IMO. That said, when it works, it works great - easy to take off just about any chip I've tried.

I've been fiddling with an old Tektronix TDS350 scope I got off ebay, it's in great shape for a nearly 20 year old machine!! It'll make debugging a LOT easier.

I have to say, getting boards made these days is a lot easier / cheaper than it was 20 years ago. I remember something like a $520 setup fee, then the cost of the boards was on top of that, $5 to $20 each or something for what we made back then (satelite tracking systems / receivers / etc).

I've done some more design work, not going to hijack your thread - will post it on my "Melissa Hands" thread.

Take Care,
Paul

[xevel]

I received the stencil today.



Here is a little idea of what it looks like. The ruler is in mm.
The size of all the apertures are closer to what they should be than what I can measure. For example, the QFN-28 should have apertures 0.25mm wide and intervals of 0.2mm, and it looks quite OK to me
The BGA has 0.3mm circular apertures aeparated by 0.2mm, very clean too.

@Paul: the oven I use is this one.
The position of the heating elements is not the best but forced convection makes it much better.

[PaulL]

Hey Xevel,

Here's what my stencils look like from Pololu (.003 inches thick). Caliper set to .039 inches (about 1mm). This is the hand board stencil (not yet used).

Left to Right (chips): HSOP8, HVQFN28, QFN24-NP(?), LQFP48.

Resistor pack at caliper is a CTS742C083, discretes are 0805. Background is some 400 grit sand paper (some debris on the sand paper, but the stencil showed up better on gray).

It's funny how certain things show up on a camera - I noticed the inner pads on the left seemed ever so slightly deformed, but I didn't see the yellowish burn on the left hand side in the pic. However, the "haze" (smoke from laser) isn't as noticeable in the pic, but you can see it drifts up and right from the holes in the pic.

Yours looks nicer, I think. (ex, no burn, no smoke marks, square inner pads):



So, how's your project coming along? I'm working on boards, they showed up Fri, got 5 done today (the easiest ones), still have the controller, power, and hand boards left (plus an interconnect board), so 5 total remaining.

Take Care,
Paul

[xevel]

Hey Paul,

Indeed, from the pictures, my stencil looks a little bit better, but there is more to the storry than just the shape of the appertures. I find that having clean edges (without blobs of fused plastic) also affect the quality of the result.
The one I have pictured does not have any, but the old ones had and I used to run a sharp blade perpendicular to the surface to scrape them off.


Other than that... not much has happened. I assembled some boards, and then had some other things to do. The competition I was developing the robot which needed these boards for has been canceled for 2014 so I turned my eyes to other stuff.
I'll get back to it whenever I feel the need for it... :/