Thanks for the detailed report. A few notes based on what's in the log:
The missing intrinsics files are not the cause of the abort — those ERROR in json_from_yaml, failed to open file / tracking_front_intrinsics.yml warnings are expected if you haven't run camera calibration yet, and voxl-camera-server falls back to default intrinsics. That's noise, not the crash.
The actual "Aborted" with no further message is what we need to isolate. The docs' Debugging Camera Issues page is the right starting point for this class of failure, but the excerpt we have doesn't show the specific abort signature you're hitting, so I don't want to guess at root cause here. Can you get us:
dmesg output right after the crash (look for a segfault/kernel oops around the same timestamp)
Whether there's a core dump generated (check /data/core* or wherever core dumps land on your image) — if so, a backtrace via gdb would tell us exactly which line aborts
Run voxl-camera-server under gdb directly (gdb --args voxl-camera-server, then run, then bt once it aborts) — this is the fastest way to get a real stack trace instead of guessing
The "works after a couple of systemctl restart calls" behavior strongly suggests a race/init-order issue rather than a config problem — since your config itself (both the custom_camera_config.txt and generated voxl-camera-server.conf) looks structurally correct for two independent ar0144 cameras. One thing worth double checking: you're using ar0144-fsin-combo on J6_LOWER_COMBO1 for tracking_down, but I don't see a J6_LOWER_COMBO1_NAME2 or combo-mode field set — since this is a combo slot, confirm whether that flex is truly running in single-sensor mode. If the combo detection is flaky on init (i.e. it doesn't always see the second physical sensor consistently), that could cause a transient failure on cold start that clears on restart.
Please also confirm which mach.var / carrier board (M0054) revision and whether tracking_down is on a combo flex physically wired for single-sensor use — a photo of the flex connection would help us rule out a physical/timing issue on that port.
