RVC4 DEV-KIT: FSYNC error with IMX582 on CSI_4 and OTA update hangs

andrea_aurea

Hi,

I'm using the following hardware and software:

Device: RVC4 DEV-KIT
SoM: OAK4 SoM
Camera: OAK-FFC-48MP (IMX582) connected to CSI_4
OS: Luxonis OS RVC4 1.22.0
Agent: 0.17.0
DepthAI: 3.8.0
oakctl (PC): 0.23.0

Issue 1: FSYNC error prevents any pipeline from starting

The camera is correctly detected:

import depthai as dai
d = dai.Device()
print(d.getConnectedCameraFeatures())

Output:

[{socket: CAM_E, sensorName: IMX582, width: 8000, height: 6000, supportedTypes: [COLOR], hasAutofocus: 0, hasAutofocusIC: 0, name: CSI_4}]

However, every pipeline that uses the camera fails to start.

For example, running the VideoEncoder example results in:

RuntimeError: RPC 'startPipeline' failed: 
Could not determine FSYNC pin for camera socket CAM_E.

Before that, the following errors are printed:

DeviceTreeDiscovery: fsync-stm-output property not found for cam_4 DeviceTreeDiscovery: fsync-stm-output property not found for cam_2 DeviceTreeDiscovery: fsync-stm-output property not found for cam_0 DeviceTreeDiscovery: fsync-stm-output property not found for cam_5 DeviceTreeDiscovery: fsync-stm-output property not found for cam_3 DeviceTreeDiscovery: fsync-stm-output property not found for cam_1

The same happens with other camera examples (for example camera_all.py).

I also checked the device tree:

grep -R "fsync" /sys/firmware/devicetree/base

which only shows the STM FSYNC node:

.../stm_fsync@12/compatible: luxonis,fsync-stm
.../stm_fsync@12/name: stm_fsync

There are no fsync-stm-output properties anywhere in the device tree.

It looks like DepthAI detects the IMX582 correctly on CAM_E (CSI_4), but cannot determine the corresponding FSYNC pin because the required device-tree properties are missing.

Issue 2: Unable to update from OS 1.22.0 to 1.35.0

Since I noticed that my device is running an older image (1.22.0), I tried updating it using oakctl.

oakctl reports that the latest available version is 1.35.0.

The connection works correctly:

oakctl connect 192.168.200.50
oakctl device info

Both commands succeed.

The board also has Internet access (ping to both IP addresses and domain names works correctly).

However, when running:

oakctl device update

I confirm the update and oakctl reports:

Firmware will be installed from: https://luxonisos-rvc4.fra1.digitaloceanspaces.com/OTA/full_update_luxonis_ext4-1.35.0.zip

Then it prints:

Starting firmware update...

but the progress remains stuck at: 0% (0s)
It stays at 0% indefinitely. If I interrupt the command with Ctrl+C, oakctl reports:

Failed to update.
Firmware update was cancelled.
Failed to wait for RPC response.

After interrupting the update, the RPC connection to the device stops responding until I reboot the board.

After rebooting, the device is still running:

OS 1.22.0
Agent 0.17.0

so the update never actually starts.

Could you please advise:

Is the missing fsync-stm-output property on RVC4 1.22.0 a known issue?
Is the OTA update hanging at 0% on RVC4 1.22.0 a known issue?
Is there a recommended way to manually flash the latest RVC4 image if OTA cannot complete?

Thank you.

Oskar-Luxonis

Hey andrea_aurea - your OS 1.22.0 is quite old; later releases include RVC4 camera-sync/FSYNC and OTA-update fixes. Please update the OS first before changing the pipeline.

Since oakctl device update is hanging, try the documented on-device update over SSH:

agentconfd update https://luxonisos-rvc4.fra1.digitaloceanspaces.com/OTA/full_update_luxonis_ext4-1.35.0.zip
reboot

If that also fails, use a full OS flash via USB3; note this wipes /data and user data: Luxonis OS update and full-flash instructions.

After updating, please retry the camera example. If the CAM_E FSYNC error remains, share the new OS version and full startup log so we can investigate the DevKit camera mapping further.

Thanks,
Oskar

andrea_aurea

Hi Oskar-Luxonis

Thank you for your previous reply.

I wanted to provide an update.

The OTA update methods did not work for me. The agentconfd update command also failed, so I eventually updated the board by performing a full flash with:

oakctl device flash https://luxonisos-rvc4.fra1.digitaloceanspaces.com/PRODUCTION/luxonis_os_rvc4-1.35.0.tar.xz

The board is now running:

Luxonis OS RVC4 1.35.0
DepthAI 3.8.0

After updating, the FSYNC issue disappeared and camera pipelines now start correctly. I verified that the video_encode.py example works correctly. It displays the live camera stream and successfully saves the encoded video to disk.

However, I have found another issue related to the IMX582 sensor.

The camera reports the following supported modes:

CameraBoardSocket.CAM_E - IMX582 (8000x6000 max) 
8000x6000 @ 10.0-18.9 fps 
4000x3000 @ 10.0-30.0 fps

I create the camera as follows:

camRgb = pipeline.create(dai.node.Camera).build( camSocket, sensorResolution=(4000, 3000), sensorFps=30.0, )

output = camRgb.requestOutput( (1920, 1080), type=dai.ImgFrame.Type.NV12, fps=30, )

The requested resolution is accepted by the API. For example, if I request an unsupported resolution such as (5000, 3000), Camera::build() correctly throws:

RuntimeError: Invalid sensor resolution specified

However, when requesting the valid 4000x3000 mode, the pipeline starts but the output indicates that the source resolution is still:

Native sensor mode in use: (8000, 6000)

and the actual encoder frame rate is only around 15 fps instead of the expected 30 fps for the 4000x3000 sensor mode.

Is this expected behaviour for the IMX582 on RVC4, or could this be a bug in the camera configuration? It seems that although the requested sensor resolution is accepted, the camera still operates in the 8000x6000 mode.

Thank you.

Oskar-Luxonis

Hey andrea_aurea - great to hear the full flash resolved the FSYNC issue.

The 4000×3000 @ 30 FPS mode is advertised and accepted by the DepthAI v3 API, so seeing the sensor still run at 8000×6000 and 15 FPS does not look expected. I’ll check this with the team.

Could you please share the full minimal script, the complete startup log showing the native sensor mode, and the output of getConnectedCameraFeatures() from the updated device? The Camera node documentation may also be useful for reference.

Thanks,
Oskar

andrea_aurea

Hi Oskar-Luxonis
Here's a minimal script that reproduces the issue:

import depthai as dai
import signal
import threading
import time

quitEvent = threading.Event()
signal.signal(signal.SIGTERM, lambda *_args: quitEvent.set())
signal.signal(signal.SIGINT, lambda *_args: quitEvent.set())

device = dai.Device()

print("getConnectedCameraFeatures():")
camSocket = None
for feature in device.getConnectedCameraFeatures(): 
    print(f" {feature.socket} - {feature.sensorName} {feature}") 
    if feature.sensorName in ("IMX582", "IMX586", "LCM48"): 
        camSocket = feature.socket

if camSocket is None: 
    print("No IMX582/IMX586 camera found on the connected device") 
    exit(1)
print(f"Using camera socket: {camSocket}")

REQUESTED_SENSOR_RES = (4000, 3000)
REQUESTED_SENSOR_FPS = 30.0

with dai.Pipeline(device) as pipeline: 
    camRgb = pipeline.create(dai.node.Camera).build( 
        camSocket, 
        sensorResolution=REQUESTED_SENSOR_RES, 
        sensorFps=REQUESTED_SENSOR_FPS, 
    ) 
    output = camRgb.requestOutput((1920, 1080), type=dai.ImgFrame.Type.NV12, fps=30) 
    outputQueue = output.createOutputQueue(maxSize=1, blocking=False) 

    pipeline.start() 
    print(f"Requested sensor mode: {REQUESTED_SENSOR_RES} @ {REQUESTED_SENSOR_FPS} fps") 

    printedSourceSize = False 
    frameCount = 0 
    windowStart = time.monotonic() 
    
    while pipeline.isRunning() and not quitEvent.is_set(): 
        frame = outputQueue.tryGet() 
        if frame is None: 
            quitEvent.wait(0.01) 
            continue 

        if not printedSourceSize: 
            print(f"Native sensor mode in use: {frame.getTransformation().getSourceSize()}") 
            printedSourceSize = True 

        frameCount += 1 
        elapsed = time.monotonic() - windowStart 
        if elapsed >= 2.0: 
            print(f"Actual output rate: {frameCount / elapsed:.1f} fps (requested {REQUESTED_SENSOR_FPS} fps)") 
            frameCount = 0 
            windowStart = time.monotonic() 

    pipeline.stop() 
    pipeline.wait()

And the full log output:

getConnectedCameraFeatures():
  CameraBoardSocket.CAM_E - IMX582 {socket: CAM_E, sensorName: IMX582, width: 8000, height: 6000, supportedTypes: [COLOR], hasAutofocus: 0, hasAutofocusIC: 0, name: CSI_4}
Using camera socket: CameraBoardSocket.CAM_E
[1346357661] [192.168.200.50] [1785741106.471] [Camera(0)] [warning] No intrinsics data found for camera on socket 4
Requested sensor mode: (4000, 3000) @ 30.0 fps
[1346357661] [192.168.200.50] [1785741106.594] [Camera(0)] [error] Wrong HW config in eeprom: ""
Native sensor mode in use: (8000, 6000)
Actual output rate: 12.2 fps (requested 30.0 fps)
Actual output rate: 15.0 fps (requested 30.0 fps)
Actual output rate: 15.0 fps (requested 30.0 fps)
Actual output rate: 15.1 fps (requested 30.0 fps)
Actual output rate: 15.1 fps (requested 30.0 fps)
Actual output rate: 15.1 fps (requested 30.0 fps)

Could either of these two issues (the missing intrinsics warning or the EEPROM HW config error) be the root cause of the sensor mode not switching?

Thank you,
Andrea

Oskar-Luxonis

Hey andrea_aurea - thanks for the detailed reproduction. The missing intrinsics warning should not determine the native sensor mode for this basic output. The empty EEPROM hardware configuration is more relevant, as it may affect how the RVC4 camera service identifies and configures the attached sensor.
Since 4000×3000 @ 30 FPS is accepted but the device actually runs at 8000×6000 and 15 FPS, this does look like an issue. Could you, before I investigate further, just update to the newest Luxonis OS?
Thanks,
Oskar