Good tip. For anyone testing without the GPS module attached, setting SYS_HAS_MAG=0 and EKF2_MAG_TYPE=5 seems to be the key part since the compass is on the GPS board. I was also looking at some compass-related references on 123compass.org while trying to understand the magnetometer side of it.

More detailed GPS documentation would definitely help, especially for troubleshooting setup issues and understanding configuration differences between modules. Saves a lot of guesswork during integration.

You might also want to look into combining depth sensing with a simple occupancy grid or voxel map instead of relying only on direct obstacle detection. It usually gives smoother avoidance behavior, especially in tighter spaces. I had better results once I started filtering noisy sensor data too.

That pink hue usually ends up being related to white balance or ISP tuning rather than the sensor itself. I’ve also seen it happen under certain lighting conditions where the auto settings overcompensate a bit.
Might be worth checking the raw output vs processed output to narrow it down. If the RAW image looks fine, then it’s probably somewhere in the color pipeline.
Also, for comparing the tint between frames, I’ve used https://colorpickerimage.org/ before to inspect the actual RGB values in different areas of the image. Makes it easier to confirm whether the pink cast is consistent or only affecting highlights/shadows.

Adding more cameras can definitely improve robustness, but synchronization and calibration become super important once you go beyond the default setup. Even small timestamp offsets between sensors can throw VIO off pretty badly. I’d also keep an eye on bandwidth and CPU usage because extra camera streams add up fast on embedded systems.

You might also want to double check the current rating of the power module versus the full load from your setup. A lot of weird reboot or brownout issues end up being power related, especially once extra peripherals get added. Keeping the wiring short and using a stable regulator helps too.

One thing that helped me before was comparing the device tree and driver setup against an already supported sensor with similar specs. A tiny mismatch in clocks or I2C config can stop the sensor from initializing properly.

If you're still having trouble getting that single sensor to initialize, double-check that your voxl-camera-server.conf file is strictly set to only enable that specific physical port. Sometimes the server tries to look for the default stereo pair and throws an error if it doesn't find both.
It also helps to run voxl-inspect-cam -a just to see if the hardware is even being detected at the system level before diving deeper into the JSON settings.

I have the same problem before — turns out the compass setup finally worked after I did a full sensor recalibration in QGroundControl and made sure the GPS/compass was facing the right direction. Worth a try if you haven’t done that yet!
You can also try this tool for free. https://123compass.org/