1
$\begingroup$

We need a SLAM system for mobile platform which will be used in on-site construction.

We are to build a 3D map from clouds of points and need localization in it.

We don’t need fast localization, but we need high accuracy of localization, error of localization should be around 1mm at a distance up to 5 meters.

Which sensor will suit for this - maybe it will be enough only a stereo vision (can it give us required accuracy?), or only LiDAR sensor (what’s about accuracy and usage in direct sunlight?) or we need to use something different and combine several type of sensors?

Can you advise the best solution for this task?

$\endgroup$
2
0
$\begingroup$

I recommend using LIDAR because stereo cameras don't work well in low lighting so if it gets shady your robot will do worst.

$\endgroup$
1
  • $\begingroup$ Mentioning one disadvantage of visual sensors seems biase to me. :)' $\endgroup$ – CroCo Feb 7 '18 at 13:26
0
$\begingroup$

First of all, performance of localization depends more on the processing power and the algorithms you have on your robot and less on the sensor you choose. Also, almost none of the SLAM algorithms and sensors you would use will give you 1mm of accuracy. Having said this here would be some options for you:

  1. Kinect : Kinect or an stereo camera can be a pretty good sensor for localization. It can give you 3D point cloud and can be used as a 2D LIDAR too. If you don't want to spend much of money, this is a good option. But again, this has limited range, FOV and can't be used in direct sunlight.

  2. Lidar: These are better than stereo vision cameras and have wider and larger scanning range. Since you'll be using this on a construction site and assuming a cluttered environment, you would need a 3D lidar. Again these are very costly and if you're just using it to just autonomously navigate through your construction site, a 2D laser scanner is cheaper and more efficient option. This is viable option and I have used it extensively without any problems.

$\endgroup$
0
$\begingroup$

A high-performance outdoors LIDAR is almost certainly going to be part of the solution. You still won't get 1mm out to 5m, though.

Another sensor I would add would be stereo vision. Kinects are kind-of bad in direct sunlight; they are indoor devices; the later RealSense sensors from Intel are better in that regard but certainly not perfect.

Also, consider that dust, sparks, and smoke in a construction site will impede the ability to localize. If you need really precise localization of the robot (rather than other objects) then you can put a number of beacons around the construction site, and let the robot localize using millimeter radio instead. Look for example at https://www.pozyx.io/

$\endgroup$

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