Ask Your Question

Which technique for 2D SLAM is easy to learn?

asked 2013-10-27 23:27:25 -0500

sarkar gravatar image

Hello Everybody,

A ROS newbie here. I'm learning ROS and currently at the file system tutorials. I'm currently running ROS on 64-bit Ubuntu 12 and soon I'm going to install it on Beaglebone Black (Angstrom).

I have a Logitech C920 webcam with hardware H264. Please suggest me 2D slam technique which can be used with my webcam and is easy to learn.

Regards, sarkar

edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted

answered 2013-10-29 11:49:47 -0500

sudhanshu_mittal gravatar image

Hi 2D SLAM gives a 2D map of the envirnment called GMAP.(Refer to GMAP gives you detailed information about the object occupancy in that 2D plane. This occupancy grid generated from gmapping gives a accurate scaled information about the actual real world. Each x and y coordinate in this 2D grid is accurate just scaled to some particular resolution, whereas in image captured from camera is a perspective image hence there is no possible transformation possible from image to real world coordinates.

2D SLAM is technique, which is quite easily be performed using lasers. Since you are a "newbie", I would suggest you to use Kinect, which is cheapest multi-sensor available.

I am not saying that it is impossible to do 2D SLAM from a webcam, but it would require a lot of knowledge and effort.

edit flag offensive delete link more


Related sub-question, sudhanshu_mittal, Is it fairly easy to implement the 3D SLAM instead of the 2D SLAM in the Navigation stack? I am wondering because I am more familiar with ROS but not yet familiar with the Navigation stack or SLAM :) Thanks!

MartinW gravatar image MartinW  ( 2013-10-29 12:04:11 -0500 )edit

Definitely 2D SLAM is much more easier to implement that 3D SLAM, I have implemented 2D SLAM with turtlebot as well as seen people implement 2D SLAM on self built robots using navigation stack. What i believe from my experience is, 3D SLAM is harder to implement and computationally very expensive.

sudhanshu_mittal gravatar image sudhanshu_mittal  ( 2013-10-29 12:20:37 -0500 )edit

sudhanshu_mittal, I have IR Proximity sensors which I can easily connect to Beaglebone Black. Are they good choice for 2D slam? I think I'd need some kind of middleware to read their data in ROS through BBB pins. sarkar

sarkar gravatar image sarkar  ( 2013-10-30 07:26:47 -0500 )edit

and the Kinect options seems reasonable. It's advantageous due to it's design as a slam and image processing sensor. Please provide a tutorial on how to use kinect for slam. However I think kinect doesn't have hardware H264 encoding that would make it difficult to use with Beaglebone.

sarkar gravatar image sarkar  ( 2013-10-30 07:29:36 -0500 )edit

Thanks for the reply! I eventually want to get 3D recognition (maybe SLAM) within the robot project I am working on but first I will get 2D SLAM working and then work on the more complex problems. Kind Regards, Martin

MartinW gravatar image MartinW  ( 2013-10-30 08:22:21 -0500 )edit

@sarkar: I have no experience with Beaglebone Black, but i can definitely say that its not possible to perform 2D slam/ gmapping from simple IR proximity sensor, since it gives only single point distance plus the data measured is raw.

sudhanshu_mittal gravatar image sudhanshu_mittal  ( 2013-10-30 14:43:38 -0500 )edit

@sarkar: First of all, if you have a ROS supported robot like turtlebot,..then you just have to launch a few files. If you are working with a self made robot, then a few extra steps need to be performed before using the inbuilt packages of ROS. Let me know, which robot you are using ?

sudhanshu_mittal gravatar image sudhanshu_mittal  ( 2013-10-30 14:48:56 -0500 )edit

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Question Tools



Asked: 2013-10-27 23:27:25 -0500

Seen: 1,549 times

Last updated: Oct 29 '13