luxonis
Problems with OAK-D-Lite running rgb_mobilenet.py
I am using OAK-D-Lite. When running rgb_mobilenet.py, the temperature of the device chip reaches 130 degrees and the program crashes after 10 minutes. Is there any solution?
Hi @2310296215 ,
Sorry about the trouble here. So on reaching 130 degrees, can you provide units on that (e.g. Fahrenheit, Celsius, Kelvin, etc.)?
And on it crashing - what host computer are you using this with. Generally crashes are a result of insufficient power from the host.
Thanks, Brandon
Host: Ubuntu 20.04 Desktop
https://github.com/luxonis/depthai-python/blob/2a0639055fe09ec4e5137087128da85c4f8c59c9/examples/SystemLogger/system_information.py#L6-L15 We use the code in the link to check the temperature,
Thanks @2310296215 We're investigating internally and will get back to you as soon as possible.
Hi @2310296215 ,
We think we figured it out. Sorry again about the trouble.
Can you please reach out to [email protected] so we can mail you a new unit? We'll be mailing you an OAK-D-S2 as a replacement (a more-capable model) as it's unclear how many OAK-D-Lite in the field (and in inventory) have the issue you are facing (more details below; we're triaging).
The Design Failure:
This comes down to a nice to have change that was done mid-production of OAK-D-Lite (we think after the first 9,400pcs were produced).
So a nice to have change was done to the enclosure to make Tripods that have extra-long screws secure into OAK-D-Lite. And this was done without checking for interference internal to the PCBA from the change. So this extension of metal into the housing caused the EEPROM in the design to be crushed by the enclosure (specifically, by the aluminum that makes up the tripod hole - which was made deeper/bigger).
And the nature of the crush is that the units work fine for some period of time. As the eeprom hasn't yet "given up the goose" in terms of letting the enclosure crush it. But eventually the eeprom "gives up the goose" and the enclosure finds it's way through the part and into metal pads internal to it.
(Note that we have not verified if all units with the outlined CE/FCC are in fact of this BAD nice-to-have design. So don't think if you have this version that your unit is for sure bad.)
And over time this crushes the eeprom, and causes the PCBA to short and thereby heatup as the power supply is providing power into a short-circuit (or close to it). Which then causes the heat that you are seeing (there's a substantial internal short).
We're still triaging how many units this impacted. So far it seems all units after the very first 9,400 that were produced. All because of one of these oh this will be nice sort of quick nice-to-have mechanical changes to the design (without design review/approval).
So we'll be replacing any/all units that have this issue, and we've already triaged internally how to prevent such a Design Failure to occur in the future (TLDR: Nice-to-haves are not Nice-to-haves, they're unqualified design changes that merit entire re-verification, re-testing, and re-control-run and field test for each of them).
Sorry about the trouble - we'll get you taken care of. And as a bonus, you'll get an OAK-D-S2 instead of an OAK-D-Lite. The OAK-D-S2 has a higher-quality RGB sensor that's also capable of higher frame-rates than the OAK-D-Lite. And it also has higher-resolution and higher-quality global shutter stereo pair (1280x800 resolution instead of 640x480 resolution).
And if anyone else runs into these problems - please email [email protected] and we'll get you taken care of.
Thanks - and sorry again about the trouble, Brandon
Thanks @Luxonis-Brandon We used an OAK-D-Lite test that didn't damage the chip and found that the temperature was normal, I'm working on Arducam, someone from our side will contact you later, thank you
Mentioning this here, as it's the "central place" of this issue. A workaround is also to grind the tripod mount metal part, so the enclosure will be able to shut completely - so TIM will be directly on the VPU chip, and there won't be issues with overheating.
This problem affected some cameras purchased before April 2022.
Please note that new OAK-D-Lite cameras have identical outlined engraving, but we have recalled all defective cameras. So if you have purchased OAK-D-Lite recently, it is not affected by the design flaw described here. However, if you encounter any issues with the camera, please get in touch with [email protected].
Mentioning this here, as it's the "central place" of this issue. A workaround is also to grind the tripod mount metal part, so the enclosure will be able to shut completely - so TIM will be directly on the VPU chip, and there won't be issues with overheating.
Many thanks! Grinding does the trick, no overheating anymore
