0
$\begingroup$

Rosanswers logo

hi. we're trying to decide what board we should use for our robot prototype. any advice/recommendation is much appreciated. we're looking for an affordable board that can handle kinect and kobuki.

our robot is fairly similar to turtlebot:

  • we'll use kinect for depth camera
  • we'll use kobuki base or build our our differential drive 4 wheel drive base.
  • we'll use have one or two arms with dynamixels.
  • we would love to have low power consumption so that the robot can operate as long as possible
  • and of course, it runs ROS

from our initial research:

  • singleboard computer (i.e. raspberry pi, beaglebone black, etc): the cost is really good but it seems like they can't handle kinect well. we haven't done extensive work with ros for ubuntu arm yet, but there seems to be some limitations vs. ros full desktop.

  • full motherboards (i.e. intel i5 or i7): they definitely can do the job, but the costs are really high. and the processing power seems to be overkilled.

  • midrange options like NUC, mini-ITX, etc: i have no idea how they perform and if they work well with kinect and kobuki

any recommendation/feedback/advice is much appreciated.


Originally posted by d on ROS Answers with karma: 121 on 2014-07-09

Post score: 0

$\endgroup$

1 Answer 1

1
$\begingroup$

Rosanswers logo

Personally, I think most of the single board computers don't have enough computing power to handle the kinect well, and they'll also be a bit slow for doing motion planning for an arm. Some of the newer quad-core, 1.5GHz+ boards might be sufficient (Odroid U3 and up, Radxa Rock, new Qualcomm boards, and probably others).

The performance of the NUC and other mini-ITX boards will be very dependent on which CPU and how much memory they have. You can get most of these boards with a wide range of CPUs, so the decisions about which CPU and which x86 form-factor to choose are not strongly correlated.

I would make the decision about which CPU and how much memory you need before deciding on a form factor. The original turltlebots managed to squeeze by with a low-powered, dual-core Intel Atom processor and 1 or 2GB of RAM. For a more complex robot, you should probably move up to an Intel i5 or an i7.

Once you have your CPU picked out, look for low-powered motherboards that support it. Personally, I think the NUC would make a great robot because some of the newer NUC support a wide-range voltage input that shouldn't need much or any extra regulation when running from a battery. This part of the NUC spec is NOT OBVIOUS, and NOT SUPPORTED BY ALL NUCs, and you have to read the full technical specification VERY CAREFULLY to find it, but it's there. Some of the NUCs also have a alternate 2-pin molex power connector on the board that will be more secure than the external barrel connector. There are also a fair number of other low-power motherboards that have single power supply inputs (mostly 12V), and the PicoPSU power supplies that are very compact and will run from a DC input.

RAM is pretty cheap these days, so buy as much as you can fit in your motherboard. Probably 4GB at the bare minimum.

If you're building a real, production robot, be prepared to buy and try several different computer configurations before you settle on the final board.


Originally posted by ahendrix with karma: 47576 on 2014-07-09

This answer was ACCEPTED on the original site

Post score: 1


Original comments

Comment by d on 2014-07-09:
thanks @ahendrix. most of our computation is done on the cloud, so locally we only need a cpu+board to handle some light-weight tasks and act as a communication device with the cloud. the heaviest local task is kinect+kobuki for navigation. do you think a mini-itx with atom e3800 + 4g ram works?

Comment by ahendrix on 2014-07-09:
I would err towards the more expensive computer for an initial prototype, so that you can get something running, and then work on optimizing the software to decrease the computing power required. If you have a very limited budget you could try to borrow a laptop with a similar CPU to test with.

Comment by ahendrix on 2014-07-09:
This of this this way: if you buy the cheaper computer and it isn't powerful enough, you've wasted time, money, and you still have to go buy something more expensive.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.