Vision-based planning on a F450 drone

[fmukwege]

Hello MRS team, I was planning on installing a depth camera to perform vision-based planning on a previously built f450 drone . I saw that you use a Intel realsense for your vision-based applications like SLAM. I have the Following questions :
• Did you use any other depth cameras than intel real sense for this drone configuration ? We plan to use a ZED2 depth camera .
• What codes already available in the MRS ecosystem could I run to make initial tests ? And is there some codes of the MRS ecosystem that won't work with this type of depth camera ?
• Is there anything I should bear in mind while using a camera with the MRS system ?

We are currently more in an explorative phase so all advice on vision-based planning from your side is more than welcome :) !

Best regards ,

Frank

[matemat13]

Sorry for the late answer, I didn't get any notification about your question :(

1. We tried some ZED camera long ago and didn't have great experience with it, but we don't have any hands-on experience with the ZED2. I know that some other teams use it, so I guess it's fine. Other than that, you can also check out the Nerian line of sensors with which we have good experience. It's well-made, offers a wider range of baselines (it can even work with custom-made stereo cameras) and has good configurability. But it's also more expensive than Realsense.
2. If the camera ROS driver outputs the depth images in a standard ROS format, then there should be no problem here. Regarding code - it depends on what you want to do. If you want to use it for reactive obstacle avoidance, use the MRS bumper. If you want to e.g. detect other drones, you can try the UAV detection algorithms.
3. Maybe I'd note that saving images (especially compressed) to a rosbag at a full rate of the camera can add significant CPU load, so watch out for that (this applies to ROS in general).

Best of luck!