with rtabmap.launch, by default it starts visual odometry node, rtabmap node and rtabmapviz node. If you just have a stereo camera and you want to do visual odometry, you may just have to set stereo:=true and remap the image topics. See this page for many stereo examples:
I have RGBD camera, 3d-lidar, rtk-GPS, imu and wheel odometer as well.
but as far as when the vehicle turns and wants to turn back to the first location, the point cloud is completely messed up and it does not provide a clear point cloud of the environment. I am using rtabamp in mapping mode, and tries to do SLAM directly, i mean, I do mapping of the environment manually, and then sending a goal in the mapped environment to the vehicle, for autonomous navigation.
I have also read in this forum that if we want to have a global loop closure, we need to use visual odometry, because of this I want to use an RGB-D camera, to resolve this issue.
If i use camera, should i deactivate icp-odometry?