bioloid daughter board for gumstix and closed-loop control

[limor]

Hi, I hope we can have an interesting discussion in this thread with the objective to end up with a good design for a Bioloid daughter board for gumstix.

This could be of benefit at least to Bullit, JonHylands, Clusher and about 8 the robocup humanoid robot league contenders.

Last week I had a chat with Clusher who is working on his Msc in robotics and will dedicate the next 6 months on perfecting a closed-loop control system for Bioloid. Clusher's boss (Prof.) said they would be willing to get such a daughter board designed and fabricated.

So what I would like to try and do is to then get RoboSavvvy and maybe other Bioloid vendors to sponsor the manufacturing of 50-100 or so boards to be sold to the Bioloid research community.



To ensure that 100 boards can be sold and meet the expectations of that many people, it is essential that several Bioloid fans are interested and actively help in the design and software of this board.

[limor]

In terms of board features, here is my take:

1) support several serial connections to independent Dynamixel busses. This is to deal the with problem of latency inherent in the Dynamixel protocol.

1.a) One way to do this is to utilize the 3 serial interfaces provided with the gumstix. But it means reducing the baud rate of the AX12 or any other dynamixel bus entity. This may deter some hobbyist from getting this board.

1.b) One interesting new MCU that can be used on this board is the ATXmega128 which is sold for 7$ and has 8 UARTS!.

1.b.a) the communication between the gumstix and the board will be through one of the UARTs using a yet-to-be-defined bi-directional compressed protocol that is suited for closed loop control at fixed rate.

2) support the different Dynamixel interfaces AX, DX, RX (means multiple port types on the board).

3) support the Robobuilder interface in order to increase the number of potential buyers for this board. Robobuilder and dynamixel are almost identical. (means yet another port type on the board).

4) interface with gumstix Verdex, and Overo

5) provide on-board support for 3-Cell LiPo batteries including low-voltage warning, cutoff and maybe even throw in a LiPo charging circuit. LiPo is something that everyone with a Humanoid wants anyway so some people may buy the board just to address the issues of charging/discharging 3C Li-Po.

[Tyberius]

All I can say is wow, this would be incredible.

Count me in for two!

I'd say the extra cost for an Atmega168 is well worth it. Having ports for AX-12s, DX/RX series, and Robobuilder is also a must in terms of appealing to the widest range of buyers. Might run this by Matt Trossen as well, he might be interested in helping.

[trey3670]

ok I am learning in this hobby half the battle is knowing what questions to ask and on that note,when you finish the product what skills will be required to use it? from what I see you are basically saying that the robobuilder software suite would replace robotis software?

[StuartL]

Matt and I aren't that far from developing our own controller board too, we were looking at one of the ARM/RISC 32 bit boards. We're aiming for something along the performance lines of the Hammer (200MHz ish) and single clock cycle instruction. We're aiming for a minimum of three 1MBit buses and interfaces for GPS, IMU etc over an SPI bus.

This is some way off but we've got home PCB manufacturing working for prototyping and can now use this to circuit test surface mount boards before production runs, reducing cost significantly.

[Robo1]

I will put my two cents in.

I've got a board with a gumstix and a 1.5M gate FPGA talking with 3 separete rs485 ports for the servo's, 3 PWM ports and some other cool things working already. With the FPGA doing all the I/O e.g. talking with the servo's and hopeflly when I finish coding it some other cool stuff. This leaves the gumstix to do the trajectury stuff. I haven't finished getting it all done but I can help you out designing a board if you want.

Will upload some pictures later today.

Bren

[Bullit]

Gumstix is about to release their OMAP3 board. It has a dual core ARM + DSP. All in the same foot print. From what I've heard its like the Verdex Pro and has the microSD right on the motherboard.
My current configuration is the Gumstix Verdex BT, microSD card, modified Gumstix breakout, FTDI2232 board, our own interface and power controller.
I have one rs485 dx/rx bus at 1Mbs and on spare high speed rs485 bus for sensors. From the Gumstix I have a 2Mbs SPI port that interfaces to my ADIS16355 IMU.

I plan on building another interface board that will include all the connectors and the FTDI 2232, the power controller, a USB hub, a 12 bit A2D which I'll put on the I2C bus, Bluetooth antenna and a battery power monitor.

Here's a photo of my current configuration for Zyn

[billyzelsnack]

Where do we find more pics of this robot!!

[billyzelsnack]

In terms of board features, here is my take:
5) provide on-board support for 3-Cell LiPo batteries including low-voltage warning, cutoff and maybe even throw in a LiPo charging circuit. LiPo is something that everyone with a Humanoid wants anyway so some people may buy the board just to address the issues of charging/discharging 3C Li-Po.

I'd stay away from the charging circuit. Seems like a big time-sink. If someone is serious about Lipos ( and people very well should be ) they should get a quality charger ( cellpro ) and never charge a battery inside their robot.

[Robo1]

billyzelsnack

I would agree with your comment, it's eaiser just to have a proper charging unit.

bren

[StuartL]

I'd stay away from the charging circuit. Seems like a big time-sink. If someone is serious about Lipos ( and people very well should be ) they should get a quality charger ( cellpro ) and never charge a battery inside their robot.

Ditto here too. Just not worth it as it's against common sense and fire-prevention.

[limor]

Tyberius: AtmegaX is a new series of MCUs from Atmel that has many improvements over the Atmega line. the important thing in this case - 8 serial ports at 1mbps. Atmega168 has 1.

StuartL: please share more about your new board project. Gumstix Verdex and now Overo (OMAP dual core, floating point processing.. at 150$) is the revolution that hobby robotics has been waiting for in that it will allow to create very stable walking/running controlled robots because it will be able to analyze the disposition of the robot and react to the environment enough times per second.

Robo1: with AtmegaX128 at under 7$ and its 8 UART ports and other assortment of standard Atmega A/D, I2C, PWM etc. it takes the fun out of fpga (at least for the purpose of I/O). And definitely appreciate help with board design.

Bullit: Where have you been hiding this great work? so you have modified one of the Verdex daughter boards? which one? and you added an FTDI in order to communicate with Dynamixel using the Verdex USB master driver hacked to lower the latency? Very impressive. What is the robot you are using and what servos? and what do you actually do on the Verdex ?

billyzelsnack: regarding charging circuit, maybe you are right that charging of LiPo is complex and dangerous and since users spend that much on their gear, it is worth getting a good charger and not risking destroying the robot. However, Sparkfun sells a Lipo charging board based on one chip.

[Bullit]

limor: The robot is of our own design. This is the third generation for us. 2nd generation with Gumstix. This new bot is fairly large at about 5kg, 75cm. It uses Robotis servos, EX, RX and DX series. All the heavy lifting is done with the EX's. The arms use a few lighter servos to reduce the weight in the upper body. The new EX's are amazing, we've tested them and found they have a stall torque of 150kg/cm at 24V!. The Verdex does all the processing at this point. We use Bluetooth for PC UI/GUI and to pilot with a PS3 DS3 controller. The Verdex does all the interval pattern generation for the servos (Beziers, s-curves etc.), Kalman filtering of IMU data, FK / IK model (still working on..) Eventually we'll be processing video also. We hope to get this robot playing soccer automously. Thus far the only drawback of the Verdex is the lack of an FPU. Floating point operations via software library are a bit slow. So far we have enough processing power but we look forward to the new OMAP processor. Nice thing about using the Gumstix is it should be pretty easy to integrate the new module right in.

If you use lipo's on your bot make sure you have voltage monitor on them!

[clusher]

We have come up with a small glitch in our plans... The gumstix serial interfaces seem to have specific objectives on the PX270 microprocessor. One is a bluetooth UART, the other is a FF UART (used to acess the console) and only one is available for everything else...

We are now evaluating the feasiblity of the board, using a slightly different approach, through the USB connection and a FTDI chip...

[srobot]

This is very interesting!

I never used a GumStix so I have a few noob questions:

1) How do you program the gumstix? (on a Win or Linux PC? Which language? etc...)
2) Which GumStix boards would I need to buy if I used this daughter board?
3) Anything else a GumStix noob should know?

Thanks,
--srobot