0
$\begingroup$

I am building an application that executes graphSLAM using datasets recorded in a simulated environment. The dataset has been produced in MRPT using the GridMapNavSimul application. To simulate the laserScans one can issue the bearing and range error standard deviation of the range finder.

Currently I am using a dataset recorded with range_noise = 0.30m, bearing_noise = 0.15deg. Am I exaggerating with these values? Could somebody provide me with typical values for these quantities? Do laser scanner manufacturers provide these values?

Thanks in advance,

$\endgroup$
1
$\begingroup$

I think 0.3m noise is a bit exaggerated for a scanning laser rangefinder. As you saw with the Hokuyo (which is one of the cheapest LIDARs you can get) they say that it is 0.03m range "error" (they do not explicitly state this is 2$\sigma$, but I have tested the noise profile myself and it is consistent with 2$\sigma$).

My experience with laser scanners is that the depth accuracy is very much a function of cost, while the angular accuracy is pretty small and is only a real problem when the lidar is used for around 10m+. A URG04LX (\$1,300) that I've calibrated had range standard deviations that were around $\sigma = 0.013m$ while a Velodyne HDL64E (>\$150,000) had around $\sigma=0.006m$. I can't remember the angular $\sigma$s but the data sheet says around 0.08 degrees (which I again assume is 3$\sigma$).

| improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ "which I again assume is 3σ" did you mean 2\sigma in this case as well? $\endgroup$ – bergercookie Jul 23 '16 at 10:39
  • $\begingroup$ @bergercookie, I use $3\sigma$ in all of my specifications, but in the case of the Hokuyo they just say "accuracy" or something. After doing some experiments I found that what they call "accuracy" is actually $2\sigma$. I assume the Velodyne is $3\sigma$ for no good reason other than it's an expensive sensor... Someone would need to do tests to see what "accuracy" on their data sheet means. $\endgroup$ – Gouda Jul 24 '16 at 2:08
1
$\begingroup$

Am I exaggerating with these values?

No your're not but you need to filter the measurements. Also, you can double check datasheet of any laser sensor for the specifications. I've written a code for Kalman filter long time ago for a laser sensor that measures a static point located at <10,10>. Let's see how the filter is working with the noise you've stated (0.3m, .15deg).

enter image description here

As you see the measurements (green) are noisy and the filter (red) is doing a great job to estimate the true value (blue) even though the measurement model is nonlinear.

enter image description here

| improve this answer | |
$\endgroup$
  • $\begingroup$ Ok, but this requires the implementation of a Kalman filter which is a totally different approach to the graphSLAM I am trying to implement isn't it? I looked up a typical laserScanner device which states that Accuracy(±30mm). If this corresponds to a gaussian model with 0 mean, this would result in \sigma = 5mm (which is far off the 0.3m that I considered) $\endgroup$ – bergercookie Jul 22 '16 at 17:59
  • $\begingroup$ I've used Kalman filter as an example to test the noise you'r mentioned regarding a noisy laser sensor. Whether Kalman is applicable or not for your project is irrelevant. Usually people assume Gaussian noise for laser sensor. $\endgroup$ – CroCo Jul 22 '16 at 18:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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