When running Monado set the environment variables VIVE_SLAM=1 and
XRT_DEBUG_GUI=1. You will need to manually press the "Submit to SLAM"
checkbox in the "SLAM Tracker" box on the debug gui.
It's not enabled by default due to some timestamps initialization problem I've
not managed to fix.
The camera orientation provided by the Index factory calibration seems to be a bit off (or we just can't get to parse it correctly). So you'll need to manually calibrate the headset if you want a better experience. Right now there are some placeholder values (that come from a calibration I did on a specific headset) that should work, but they will likely be subpar for your setup.
Also, remember we have community-made drivers, we are not sure whether the timestamps we are using are correct, so that might be another source of inaccuracy.
All in all, it is still usable. Try it out, it should work out of the box, and only if you want a better experience, proceed with the calibration procedure explained below.
The most annoying bit is that you will need to manually calibrate the headset to
get a json file for Basalt like the index_calib.json
example. You can use that example first to see how it works for you by setting
SLAM_CONFIG=$bsltdeps/basalt/data/monado/index.toml, but that will likely not
work very well.
The general calibration procedure is as follows:
- Setup the calibration target:
- Download the one from kalibr
- Open the pdf file and measure with a ruler the black square side (e.g., 24 inch 1080p screen, 34% zoom in Okular viewer, gives 3cm)
- Update aprlgrid_6x6.json "tagSize" property in meters (e.g., 3cm would be "0.03")
- Record camera calibration sequence with the euroc recorder in Monado:
- See calib-cam3 example from TUM-VI https://vision.in.tum.de/data/datasets/visual-inertial-dataset
- Or this example from ORB-SLAM3: https://www.youtube.com/watch?v=R_K9-O4ool8
- Note that for stereo calibration you want to cover as much as possible of both cameras images but always trying to keep >90% of the calibration target visible in both views. This can be a little tricky so practice it beforehand.
- Record camera-imu calibration sequence: (faster motions)
- Similar recommendations as in the previous sequence but this time we want faster motions that excite all IMU axes, see these examples:
- calib-imu1 from TUM-VI https://vision.in.tum.de/data/datasets/visual-inertial-dataset
- Or this from ORB-SLAM3 https://www.youtube.com/watch?v=4XkivVLw5k4
- Now run camera calibration as explained here
- Then camera+imu (no mocap) calibration as explained here
- If the datasets are not good, the calibration will likely not be very good as well. If you want to be sure you calibrated things properly, a good idea is to repeat this process until you get similar results.