SLAM with two d455 sensors

classic Classic list List threaded Threaded
9 messages Options
Reply | Threaded
Open this post in threaded view
|

SLAM with two d455 sensors

dinossht
Can anyone help with running two d455 sensors? They are mounted back to back, 180 degrees oriented. Is it correct that the following should be set in odometry as such for multiple sensors strategy to work?

   
param name="Odom/Strategy"            type="string" value="0"
param name="OdomF2M/BundleAdjustment" type="string" value="0"
param name="Vis/EstimationType"      type="string" value="0"

Currently, the system odometry is struggling..
Reply | Threaded
Open this post in threaded view
|

Re: SLAM with two d455 sensors

dinossht
This post was updated on .
My current launch file looks like this

Reply | Threaded
Open this post in threaded view
|

Re: SLAM with two d455 sensors

dinossht
and

Reply | Threaded
Open this post in threaded view
|

Re: SLAM with two d455 sensors

matlabbe
Administrator
That is an old example (that was with kinect 360?), if rtabmap is built with OpenGV, I recommend to use default visual odometry settings:
Odom/Strategy=0
OdomF2M/BundleAdjustment=1
Vis/EstimationType = 1
It will be a lot better.

I would not set  Vis/MaxDepth=4, keep it to default 0.

cheers,
Mathieu
Reply | Threaded
Open this post in threaded view
|

Re: SLAM with two d455 sensors

dinossht
Yes, the odometry is  bad with 3D->3D. So I will try the default settings as you recommend. I am struggling to install rtab-map with opengv. Do you have any Docker files I can test?
Reply | Threaded
Open this post in threaded view
|

Re: SLAM with two d455 sensors

matlabbe
Administrator
introlab3it/rtabmap_ros:noetic-latest

This container contains rtabmap built with opengv.
Reply | Threaded
Open this post in threaded view
|

Re: SLAM with two d455 sensors

dinossht
Reply | Threaded
Open this post in threaded view
|

Re: SLAM with two d455 sensors

dinossht
So I managed to run with opengv. But it seems like the visual odometry performs better with single depth camera rather than two depth cameras. My hope was to improve the odometry and SLAM stability (meaning less odom loss) by adding two d455 camera, each of them facing on the opposite direction for a large FOV. Does this assumption make sense if I run rtabmap with multi camera mode?
Reply | Threaded
Open this post in threaded view
|

Re: SLAM with two d455 sensors

matlabbe
Administrator
Yes, this is that dockerfile.

I didn't do exhaustive accuracy comparison between single and multi cameras setup, but the latest can be a lot more robust to occlusion/textureless environment. To work correctly, the TF between the two cameras should be pretty accurate. Note also that if the cameras are poorly synchronized, it won't work.

cheers,
Mathieu