Skip to main content

Use Arducam OV9281 (Jetvariety) Camera with A603 Jetson Carrier Board

Arducam Jetvariety cameras officially support the NVIDIA developer kit carrier board, not the A603, and Arducam's installer overlay cannot work on the A603. The camera itself works well once you build a small custom device tree. This page documents the working configuration for the Arducam OV9281 (B0223) on JetPack 6.2 (L4T R36.4.3), the reasons the stock overlay fails, and one trap that can disable all USB on the board.

Result: OV9281 on /dev/video0, GREY / Y10 / Y16 formats at 1280x800, 1280x720 and 640x400, up to 80 fps, with external trigger available.

Why Arducam's own overlay cannot work on the A603​

  1. The A603's single CSI connector is serial_a on i2c-2 (i2c@3180000). Arducam's developer-kit overlay wires its sensor nodes to serial_b/serial_c under a cam_i2cmux node, a mux that never instantiates on this board, so the sensor nodes never probe. There is no error message; the camera simply never appears.
  2. The Jetvariety controller answers at i2c address 0x0c. Verify with i2cdetect -y -r 2.
  3. The A603 BSP ships its own rebuilt kernel and reuses NVIDIA's device tree file names, so the tree that must be modified is the A603 one, not a developer-kit one.

Boot safety first​

caution

Arducam's installer adds a boot entry (commonly labelled JetsonIO) pairing their kernel with their developer-kit device tree. Booting that entry on the A603 disables every USB port, including the USB gadget network and any keyboard. The A603's UEFI/extlinux boot menu accepts no keyboard input, so whatever DEFAULT points at in /boot/extlinux/extlinux.conf is what boots, always. Never leave DEFAULT pointing at an untested entry. Recovery from this state requires RCM-booting the flashing initrd from another machine (recovery jumper on W7 pins 3 and 4).

Before experimenting, back up the original kernel and DTB, and consider installing a systemd timer that resets DEFAULT to a known-good entry about 180 seconds after every boot unless a keep-flag file exists:

# /usr/local/sbin/a603-boot-revert.sh
#!/bin/sh
[ -e /etc/a603-keep-boot-default ] && exit 0
sed -i 's/^DEFAULT .*/DEFAULT seeed/' /boot/extlinux/extlinux.conf

Workflow for any boot experiment: remove the keep-flag, point DEFAULT at the experimental entry and reboot. If it works, create the keep-flag again. If the board becomes unreachable, power cycle and it reverts to the known-good entry on its own.

The working configuration​

Boot entry in /boot/extlinux/extlinux.conf:

LABEL arducam
      MENU LABEL Arducam kernel + A603 sensor-swapped DTB
      LINUX /boot/arducam/Image
      FDT /boot/kernel_a603-arducam.dtb
      APPEND ${cbootargs} root=/dev/nvme0n1p1 rw rootwait rootfstype=ext4 ...
  • Kernel: Arducam's, from their arducam-nvidia-l4t-kernel package (their installer fetches it). It carries the arducam-csi2 Jetvariety driver.
  • DTB: custom, built as described below. Note that the FDT line in extlinux does override the partition DTB on this board.

Device tree strategy: swap the sensor, keep everything else​

Grafting Arducam's camera graph into the A603 tree fails on phandle and wiring mismatches. What works is the opposite: keep the A603's complete, board-correct camera graph (CSI channel, VI wiring, ports and endpoints, phandles, all built for the IMX219 the board supports out of the box) and swap only the sensor node identity:

  1. Decompile the stock A603 DTB: dtc -I dtb -O dts -o seeed.dts <stock DTB>
  2. Decompile any DTB that contains Arducam's sensor block (for example from their overlay) to obtain its mode0 node
  3. In seeed.dts, inside i2c@3180000, replace the rbpcv2_imx219_a@10 node with an arducam_a@0c node:
    • compatible = "arducam,arducam-csi2", reg = <0x0c>, sensor_model = "arducam-csi2", use_sensor_mode_id = "true"
    • keep the original devnode, reset-gpios, phandle and the entire ports block verbatim; this keeps every phandle valid
    • insert Arducam's mode0 block with tegra_sinterface changed to "serial_a"
  4. Flip the cam0-rst GPIO hog from output-low to output-high (the stock tree holds the camera in reset)
  5. Update the tegra-camera-platform path strings and __symbols__ from .../rbpcv2_imx219_a@10 to .../arducam_a@0c
  6. Recompile and install: dtc -I dts -O dtb -o /boot/kernel_a603-arducam.dtb a603-arducam.dts

A Python script that performs this transformation automatically is available in this repository.

Success signature in dmesg:

arducam-csi2 2-000c: firmware version: 0x10003
arducam-csi2 2-000c: Sensor ID: 0x9281
arducam-csi2 2-000c: sensor arducam-csi2 2-000c registered

Check the ribbon cable before debugging software​

If the sensor does not answer on i2c, check the physical layer first:

  • At the camera end, the ribbon's gold contacts must face away from the camera PCB, and both latches must be fully and evenly closed.
  • A decisive diagnostic is the free-end power test: seat the ribbon in the A603 only, leave the camera end loose, power the board, and measure the loose end's outermost contacts with a multimeter. Reading 3.3 V proves the board, the cable and the board-end orientation are all good, isolating any remaining fault to the camera end.
  • Cable: 22-pin, type A (gold on the same face at both ends), straight-through.

Capture​

note

The driver's default exposure and gain produce black frames, which looks exactly like a broken camera. Set controls in the same v4l2-ctl invocation as the capture, because values reset every time a new stream opens.

v4l2-ctl -d /dev/video0 --set-fmt-video=width=1280,height=800,pixelformat=GREY \
  --set-ctrl exposure=10000 --set-ctrl analogue_gain=400 \
  --stream-mmap --stream-count=10 --stream-to=/tmp/frames.raw

Ranges: exposure 1 to 65523, analogue_gain 100 to 1500. Also available: trigger_mode (external trigger), frame_rate 5 to 80, horizontal and vertical flip.

While the approach was tested with the OV9281 B0223, it should apply to other Jetvariety modules on the A603, since the Jetvariety design puts the mode logic in the camera's onboard controller rather than the device tree.

Loading Comments...