I've worked quite a bit with these issues, and I would say that you'd be surprised how extremely efficient ROS is. I must say though, what used to be very clear comparison between scalar numbers has become something that tells me nothing. 2.8Ghz, that's more than the Intel Joule's 2.4Ghz-bursts? RIght? I had one Joule doing SLAM, interface the controller, rgbd with computationally cheap 3D-camera, running high-res 2D laserscan and even running my code (vastly inefficient) for getting some data from another sensor through multiple nodes. I can't say it was a breeze, because the Joule seems to actually have passed right after - but it managed. Keep in mind that I was developing at the same time, starting and stopping things, looking at the nice visualizations (on diff computer ofc) and other stressors for the poor guy. This happened earlier today, and now I've moved the majority of it to an Odroid XU-something (fancy blue fan). *EDIT: The Odroid is not able to process 3D in parallel to the other stuff, won't even bother to try. *
I would say you can do the task you describe, given the right parameters (dunno how an exact amount of computational expense can be given without), but it will be a pain to set up and to work on. Would recommend going through everything needed beforehand on workstation, fixing all bugs and setting up a script to do the building and downloading for you. It's really, really painful to ssh through the error messages and compiler-mysteries hours on end (spoiler: you're probably out of memory)).
They may not look like it, but even RPis and Intel Edisons are computers. I've updated your question title to -- what I think -- more accurately reflects the contents of your question.
yes I am aware of that, raspberry pi 3 is a mini computer compared with nowadays computers with 1.2GHz but how much time does it need if it works at all, do I need almost 5 seconds for full loop, and is there a better alternative?
FYI it was recently announced that the Intel Edison would be discontinued.
I suggest to test it for your use case.The processing power and memory for gmapping e.g. highly depends on size of your map, resolution, number of scans per second, update counts..
that is correct, thank you for reminding me. the number of tree particles will be much less for 5x5 m2 area. so for such an area, I believe raspberry pi 3 would be more than enough. but for scanning a whole house 160 m2 for example, it would pull it off ,still need further testing.