Currently, a large part of processing time for each rtabmap update is the compression of debugging data (RGB+depth+lidar data) and the compression of the local occupancy grid. There could be then an advantage to integrate nvCOMP to transfer that burden to GPU (in particular on jetson), which seems compatible with CPU zlib (so that the map can still be analysed or used by another computer not having nvidia gpu). See examples here: https://github.com/NVIDIA/CUDALibrarySamples/tree/master/nvCOMP/examples
Currently, a large part of processing time for each rtabmap update is the compression of debugging data (RGB+depth+lidar data) and the compression of the local occupancy grid. There could be then an advantage to integrate nvCOMP to transfer that burden to GPU (in particular on jetson), which seems compatible with CPU zlib (so that the map can still be analysed or used by another computer not having nvidia gpu). See examples here: https://github.com/NVIDIA/CUDALibrarySamples/tree/master/nvCOMP/examples