very simple question as anticipated in the title.
I used rtabmap on a mobile tablet equipped with T265 + L515 with amazing results and low costs.
Now that RealSense decided to dismiss both my sensors, I've searched a lot for replacements.
While it was not so difficult to replace L515 with a K4A, almost same final quality, I'm facing several problems in finding a valid alternative to T265, not in terms of performances, but relatively to the final cost of the solution.
Zed2 has everything necessary to work as an odometry sensor, but the requirements of an NVIDIA GPU is very frustrating, especially in terms of costs.
Is it possibile to run ZED SDK for odometer purposes only without CUDA?
Do you now alternatives with no such requirement?
If you want something very portable and convenient, I would say iPad Pro (models from 2020 with LiDAR). The iPad is less expensive than iPhone12-13 Pro. RTAB-Map for iOS is available on AppStore, but there are also many scanning apps there to match the application you need.
K4A is great alone, you may use rtabmap for odometry too (and even in pitch black without lights). However, if you need the robustness of a tightly coupled VIO approach, there are indeed less choices. Are you using the RTAB-Map standalone app? or with ROS? With ROS you could find a VIO approach that could work with K4A to feed to rtabmap. There are also some VIO approaches that can be built in RTAB-Map standalone that could work with K4A (though not tested myself).
For ZED, I think CUDA is required to get the pose, but you could ask them on their github. They have an open driver without CUDA, but we can get only the raw sensor data.