Ask Your Question
0

How to choose a Lidar for indoor SLAM?

asked 2018-09-08 02:15:09 -0500

simk0024 gravatar image

Hi all,

I am new to ROS and would like to start my journey from learning and building SLAM for indoor UGV. Although there are some off-the-shelf platforms that package with all the components and sensors, still I would like to ask: how to choose a suitable Lidar? There are dozen of questions come to my mind when trying to choose a lidar.

There are various kinds of Lidar with different brand, cost, specs like 2/3D, channel, distance, accuracy, angular resolution, scan speed, etc. But how to decide which is the right one or combination with best performance?

Is Lidar affecting the performance of SLAM? or SLAM algorithm could cover the deficiency of hardware? Does Lidar with better resolution, scan speed, etc requires higher performance of CPU/GPU? Is there any method to evaluate the Lidar?

I know there is no correct answer for these questions, but your comments, thoughts, experiences will definitely help me and people like me a lot!

PS. I am not sure if I post this question in the right place, please let me know if I am not.

Thank you.

edit retag flag offensive close merge delete

Comments

If you are starting with lidars then the LDS-O1 from robotis (2D lidar) is a good option and for 3D lidars I heard about velodyne (but I'm not familiar with).

Delb gravatar image Delb  ( 2018-09-10 01:52:03 -0500 )edit

For the SLAM, except for HECTOR, you don't rely only on the lidar data but also odometry and IMU. So yes having a bad lidar will affect the performances of the slam but having the best one with a bad IMU will also affect the performances.

Delb gravatar image Delb  ( 2018-09-10 01:54:17 -0500 )edit

1 Answer

Sort by ยป oldest newest most voted
0

answered 2018-09-09 07:01:46 -0500

snowee gravatar image

In my experience, this will come down mainly to price and skill.

3d lidars are vastly more expensive. 2d laser scanners aren't cheap by any stretch of the imagination, but i did see some japanese ones starting at 700$ (don't know the name unfortunately). Also, 3d slam is a lot harder to implement correctly - and efficiently - than SLAM with a 2d lidar like SICK.

If you only want to learn, scanning distance, accuracy and speed shouldn't really be your concern. Modern lidars are very fast in general, and accuracy shouldn't be that big of an issue. The resulting map may be a little imprecise, but the algorithm you'll write will work with any other Laserscanner supported by ROS. So you can always upgrade.

What kind of computer will you be running? By default ROS handles everything on the CPU. Handling laserdata on the GPU should be possible, but implementing anything really in e.g. CUDA is no easy task.

Just my two cents

edit flag offensive delete link more

Comments

Thank you @snowee. I think the Japanese brand you mentioned is Hokuyo? it's cheaper compare to SICK, but not sure abt the performance compare. For simplicity, I am target to use a Intel-i7 NUC. Lidar may cause the imprecise of map, possible that different algorithm can improve the precision?

simk0024 gravatar image simk0024  ( 2018-09-09 23:12:07 -0500 )edit

The NUC can definitely handle that. We have multiple SICK sensors connected to it in our lab, and we haven't had a single issue with performance.

Well, there are multiple variants of SLAM. They all use some form of Kalman Filtering, which should filter out inconsistencies pretty well!

snowee gravatar image snowee  ( 2018-09-10 04:08:04 -0500 )edit

Why are you worried about precision? There are people doing SLAM with cameras only, and the results are still amazing, even though estimating 3D information from vision only is really hard.

snowee gravatar image snowee  ( 2018-09-10 04:10:43 -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

2 followers

Stats

Asked: 2018-09-08 02:15:09 -0500

Seen: 437 times

Last updated: Sep 09 '18