So I have decided to test the SLAM with ZED M using the parameters you have suggested on that post. Before doing the SLAM, I wanted to ask some questions:
1. Would you recommend to use ZED M depth estimation and send rgbd_d to rtabmap or just send stereo images and let rtabmap do the depth estimation?
2. I have used these parameters and can you say whether I made a good choice? I have read the parameters one by one but thought you might suggest something (I am using the camera for SLAM and also wheel odometry):
Re: Best parameters for warehouse SLAM using ZED M
1. It depends. If you look for accurate 3D map indoor, you may use stereo mode as rtabmap will do depth estimation only where there is a lot of texture (on edges). Using ZED-M depth can be useful to not have to estimate it on CPU as it would be computed on GPU.
2. For RGBD/OptimizeMaxError, it should be >1, default is 3. The "0.1" may be referring to old way that parameter was used. Now it is a relative ratio against link's covariance instead of absolute distance. "Vis/InlierDistance" is ignored here, as "Vis/EstimationType=1" by default. "Kp/MaxDepth" and "Vis/MaxDepth" are quite small, default is 0 (infinity). Set "Rtabmap/TimeThr" to 0 if you don't want memory management behavior (maybe easier to debug SLAM before going into the memory management problem).