0
$\begingroup$

Intro: I am a student that just started a project regarding prototyping a mobile robot that involve Indoor SLAM implementation. I am very new to the field of computer vision, SLAM and SBC (single board computer).

I am searching for advice on choices of SBC for visual ORB-SLAM implementation.

Below are the few options I came across, welcome to provide new suggestion:

  1. Raspberry Pi : Is the computational power of Raspberry Pi able to support ORB-SLAM? If so, which model would be my best choice?

  2. BeagleBone Black : Any comments on this SBC for ORB-SLAM implementation?

My budget for SBC is around $150

I would also like to know a rough estimate of the minimum requirement for SBC to implement ORB-SLAM.

I really appreciate any advice, resource or help I could get.

Thanks in advance.

| improve this question | |
$\endgroup$
0
$\begingroup$

As FourierFlux mentioned, the Jetson Nano is perhaps the most processing power you're going to get for $150 and under. However getting ORB-SLAM to run in it might be a challenge — there is a port but it doesn't seem to be very actively maintained.

If splurging a little more on the SBC is an option, you could get a UDOO Bolt, which is a full x86-64 SBC built around the AMD Ryzen Embedded chipset. That would enable running the stock ORB-SLAM system. It sells online for \$332.00 (Vega 3 graphics) or \$418.00 (Vega 8).

Alternatively, if you don't require running all software on the robot itself, you could get a Raspberry Pi, set it up to stream visual data to a more powerful remote machine (e.g. a desktop PC), and run ORB-SLAM there instead. This could be easily accomplished in ROS by running hardware driver nodes on the Raspberry Pi, more demanding SLAM and navigation nodes on the remote machine, and have the two sides communicate over wi-fi:

enter image description here

See this report for details on how this could be done.

Finally, you could try a more economical SLAM approach: a successful implementation of FastSLAM2.0 for Raspberry Pi has been reported, you could contact the authors and ask for the code, or re-implement it yourself.

| improve this answer | |
$\endgroup$
  • $\begingroup$ Your input is much appreciated, contain so much information that I am seeking for. I have some other questions regarding this manner. What do you mean if I dont run the SLAM code itself on the robot I can go for raspberry pis? Whats the function of remote function (why it can overcome the lack of processing power of raspberry pis? and could I ultilize remote machine with Jetson Nano too? Sorry for asking more questions. Thanks! $\endgroup$ – Denzer Apr 22 at 19:23
  • $\begingroup$ I added a diagram illustrating what I mean, hope it helps. You could of course use a Jetson Nano instead of a Raspberry Pi in a similar setup; however the Nano's appeal is precisely packing enough computing power to run everything locally, so I'm not sure what the point would be. $\endgroup$ – xperroni Apr 22 at 20:42
  • $\begingroup$ Thank you so much for the diagram. I am glad that I got your advice :) $\endgroup$ – Denzer Apr 25 at 18:32
0
$\begingroup$

Tbh ur going to need a Jetson nano at least, maybe a tx.

| improve this answer | |
$\endgroup$
  • $\begingroup$ Thank you. A lot of people suggested Jetson Nano to me. Glad to hear from you too! $\endgroup$ – Denzer Apr 21 at 9:26
  • $\begingroup$ Welcome to Robotics FourierFlux. Thanks for your answer but we are looking for comprehensive answers that provide some explanation and context. Very short answers cannot do this, so please edit your answer to explain why it is right, ideally with citations. Answers that don't include explanations may be removed. See How to Answer for more info. $\endgroup$ – Mark Booth♦ Apr 21 at 19:50

Not the answer you're looking for? Browse other questions tagged or ask your own question.