3D visionTroubleshooting

3D camera stuck in firmware recovery: reboot to productive mode

Your 3D camera is stuck in recovery after a firmware update: read the install log, finish the reboot to productive mode, match tool versions, then decide.

LumaSpec Vision10 min read Share on LinkedIn

On this page 8 sections
  1. 01What "stuck in recovery" looks like
  2. 02Read the install log before you reflash
  3. 03Finish the reboot from the recovery web interface
  4. 04Match the SDK and viewer to the new firmware
  5. 05Rule out power and the cable first
  6. 06When it is the flash: the send-back decision
  7. 07Prevent the loop on the next update
  8. 08Questions

The update ran to the end and your 3D camera still boots into its recovery image. The command-line tool reports "Timed out waiting for productive mode", the desktop viewer returns a reboot failure, and every reconnect attempt dies with a dead-session error. In most cases nothing is broken: the installer log shows every image written, and the only missing step is the reboot command itself. This guide gives you the sequence that separates a waiting camera from a damaged one: decode the install log, trigger the reboot from the recovery web interface on port 8080, match your SDK and viewer to the new firmware, rule out power and cabling, and decide when the unit genuinely goes back. Anyone who updates 3D sensors on a production line will use this sequence more than once.

What "stuck in recovery" looks like

You know the failure by its signature. The update tool writes the package, then the CLI prints "Timed out waiting for productive mode" and exits. Rebooting to productive mode from the desktop viewer stalls and returns "System reboot failed. 120004". A device search lists the camera, but connecting fails with "Could not find a compatible device at the given IP address. 120015". The API client fails at startup with "XMLRPC object not found (D-Bus error 22)" and the phrase "trying to access dead session". Meanwhile the recovery page — at 192.168.0.69:8080 in the reported case — still loads and shows the installer log.

None of these messages says the flash write failed. They are timing and version signals first, hardware signals last. Read them against this table:

What you see What it means
"SWUPDATE successful!" in the recovery log Every image reached its flash volume; the install phase is complete
"Timed out waiting for productive mode" The tool's fixed polling window closed before the post-install work and reboot finished — not a verdict on the flash
"System reboot failed. 120004" The viewer asked for a productive boot and the camera did not return as productive inside the wait
"Could not find a compatible device ... 120015" "Compatible" is the operative word: the search saw the camera, but the version pairing or the mode blocked the connection
Port 8080 answers, port 80 does not The camera is in recovery; the productive API lives on port 80

If your camera never leaves update mode at all — no recovery web page anywhere — that is a different failure with its own first step, and the guide to a camera stuck in update mode starts with reading the USB ID.

Read the install log before you reflash

The recovery page shows the full installer log, and it is the most valuable screen in this failure. Read it from the bottom. The reported case shows three images written into named UBI volumes: the kernel into /dev/ubi0_0, the root filesystem into /dev/ubi1_0, and the application volume into /dev/ubi1_1. After the volumes, the log moves to the diagnostic stage: "DIAGCTL update SUCCESS!", then "Installation takes a while ..." with the diagnostic file still installing. That tail runs after the images are in place, and it is the slow part.

"SWUPDATE successful!" is the decision point. It means the writes are done. Re-flashing the same package again adds nothing and adds risk: every extra write cycle through a marginal link is a fresh chance at an interrupted session.

While you are in the log, note the version on the application-volume line. In the reported case it reads 1.71.6, and the packages attempted were 1.71.9079 and 1.71.7444 — the same 1.71 family. When the family matches on both lines, the install is coherent; a mismatch points at a partial or wrong package.

Finish the reboot from the recovery web interface

The missing step in nearly every one of these loops is small: the install finished and nobody told the camera to boot the new image. The recovery page has the control, labeled "Reboot to productive mode".

  1. Put the host and the camera on the same subnet, ideally connected directly with a short cable.
  2. Open the recovery web page at the camera's address on port 8080. If it loads, recovery is alive — that page is your lifeline for everything that follows.
  3. Read the log to the end. If it still shows "Installation in progress", wait; do not restart the install and do not cut power.
  4. When the log ends with "SWUPDATE successful!", select "Reboot to productive mode".
  5. Wait and poll rather than trusting the tool's timeout. Give the camera up to five minutes; the CLI's polling window is its own fixed wait and often expires first.
  6. Test which mode you are in. The productive API answers on port 80; recovery stays on 8080. Port 80 alive while the client still fails is a version problem — next section.
  7. If the camera boots back into recovery, do one clean full power-cycle — off, ten seconds, on — and run steps 2 through 6 once more. Two clean attempts are the budget.

A camera that was unreachable through the API five minutes ago and answers on 8080 now is not dead; it is late. If the camera hangs mid-sequence rather than looping back into recovery, the host PC is the usual culprit, and the write-up on a camera that hangs after a firmware update covers that branch. Connecting from a different PC than the one that ran the update can also fail for host-side reasons; that case has its own guide.

Match the SDK and viewer to the new firmware

An update moves the camera to a new firmware family, and the remote interface changes between families. An older client then produces exactly the dead-session signature: it reaches the camera, asks for an object that no longer exists where it expects one, and prints "XMLRPC object not found (D-Bus error 22)". In the reported case, client versions 1.0.1 and 1.5.1 were tried against firmware 1.71.9079 and 1.71.7444. None of those pairings matched, so every reconnect failed while the recovery page worked perfectly.

The rule is to read the paired client version from the firmware's release notes and install that exact SDK and viewer version before you reconnect — before, not after. The 120015 message, "Could not find a compatible device", is a version statement as much as a network statement.

Two live articles cover this from other angles: firmware and SDK version mismatch, check versions first, and SDK v2 firmware mismatch with no device found. The same discipline applies to code you deploy yourself — matching SDK and firmware versions on a smart camera walks that case.

Rule out power and the cable first

Between the first and second attempt, change the physical layer, because a brown-out or a flapping link mid-write is what turns a version problem into hardware damage. Bench the camera: host and camera on one short, known-good camera data cable with no switch in the path. If the camera is fed over PoE, use a port that delivers the full class budget and watch for sag under load — the mechanism behind PoE droop during exposure applies to update writes too. If it runs from a DC rail, meter the rail at the camera connector during boot and during the write; the DC line fuse and noise filter guide shows the filtering a vision rail needs.

The 3 m GigE camera cable is what we keep on the bench for update work: short enough to rule out link margins, flexible enough to move the camera around the desk.

Cables3 m GigE camera cable, high-flex (drag chain), screw locks (LS-CBL-0199)LS-CBL-0199$54.00View partBuild a system with it

For a camera on a DC rail, a regulated enclosed supply gives you a rail you can meter. The 48 V, 76.8 W industrial supply is the example from our power supplies page.

Power suppliesMEAN WELL LRS-75-48 Industrial Power Supply, 48 V 76.8 WLRS-75-48
Type
Power cords and supplies
$35.00View part

When it is the flash: the send-back decision

Escalate to "hardware problem" only after all of these are true:

  • The install log reached "SWUPDATE successful!" twice, with the diagnostic tail completed each time.
  • "Reboot to productive mode" was selected on the recovery page each time, not only requested from the viewer.
  • The SDK and viewer versions are the ones the firmware release notes pair with the image.
  • Power is metered or substituted known-good, on a short cable.

If the recovery page still loads after two clean attempts and productive mode never comes up, the productive boot chain is damaged. Stop re-flashing — repeated installs through a marginal link are how a recoverable unit becomes scrap — and arrange service through the camera vendor. If replacement is the realistic path, the 3D camera finder filters laser profilers, stereo cameras and volume cameras by type.

Do not open the housing or attempt a low-level bootloader flash unless the vendor publishes that procedure for your model; where DFU-style recovery exists, follow it step by step — DFU firmware recovery, done safely sets out the precautions.

The same install-finished-but-won't-boot pattern shows up beyond cameras. If a PLC or panel in the same cell sits in boot mode after its own firmware update, the PLC firmware recovery guide walks that device back to a running image.

Prevent the loop on the next update

Every item below exists because its absence caused a support case:

  • Record the firmware version before you update. The reported case never knew its starting version, and that gap cost a round-trip with support.
  • Install the paired SDK and viewer versions before the update, on the bench host, not after it on the production PC.
  • Update on the bench: short cable, dedicated power, nothing else on the link.
  • Read the log to "SWUPDATE successful!" and let the diagnostic tail finish before anything else happens.
  • Poll port 80 for up to five minutes instead of trusting the tool's timeout.
  • Budget two clean attempts, and write down what each one showed.
  • Write the recovery address — the camera's IP on port 8080 — into the station documentation. That page is the recovery path if an update ever goes sideways.

If the camera's address will not come back after a successful recovery, persistent IP addresses that fail to survive a reboot is the follow-up read, and the wider set of version-matching and recovery cases lives on the 3D camera topic page.

For a new station, the configurator walks the camera, cable and power list in order, or contact us about a camera bought from us — every order includes free expert support.

Questions

Can I run a 3D camera firmware update over Wi-Fi or through the office switch?

The protocol will allow it, but do not. The install tolerates network latency; it does not tolerate a dropped session. A managed switch that renegotiates a port, a Wi-Fi client that roams, or a marginal long run in the middle of a write is how a routine update becomes a recovery loop. Bench it: host and camera on one short cable, dedicated power, nothing else on the link.

Why does the API report a dead session when the install clearly succeeded?

Because the client and the new firmware no longer speak the same RPC dialect. Firmware family changes alter the remote interface, so an older SDK asks for an object that is no longer there and prints "XMLRPC object not found (D-Bus error 22)". In the reported case, clients 1.0.1 and 1.5.1 both failed against a 1.71 image. Install the paired SDK and viewer before you reconnect.

Does a failed reboot to productive mode mean the flash is corrupt?

Rarely on the first failure. Most loops end at the missing reboot command, the tool versions or the power feed. Treat the boot chain as damaged only after two clean attempts: completed install, reboot selected on the 8080 page, matched SDK and viewer, metered supply on a short cable. A recovery page that stops loading, or a camera that vanishes from the network mid-install, is the hardware-leaning signature — record exactly which stage died.

Is it safe to pull power during an update and start over?

Not while the recovery log shows "Installation in progress". The diagnostic stage runs after the main images are written, and cutting power there is the classic route from a recoverable camera to a repair. Let the log reach "SWUPDATE successful!", trigger the reboot, and only power-cycle between completed attempts. Interrupted writes, not failed boots, are what turn a version problem into hardware damage.