Modify the BW1098OBC to add an ESP32 System on Module

sanjith

Such a great work producing a minimalistic hardware like this ! kudos for that. Whats bothering my mind is that how is ESP32 talking to the MOVIDIOUS ( i understand its through SPI ), I'm more worried abt the software part of it. Are you using OpenVino in order to control movidious? if so then where would be this OpenVino installed?

thank you in advance 😁

Brandon

Hi sanjith ,

Thanks for the kind words! So the firmware to support this we are implementing directly. The neural models will still be fully OpenVINO compatible... and upgradeable via SPI/etc. But no computer is needed for it to run... in fact once the SPI protocol is fully running you will even be able to transfer the OpenVINO model over SPI completely without a computer... for example from AWS-IoT.

In fact we have an initial implementation for the DepthAI SPI communication (streaming of NN metadata for now). The more final SPI protocol is not integrated yet (we are building it internally, will be CRC-checked binary format), so for now the packet consists of a NULL-terminated string.
The DepthAI boot and initialization is still done by USB, and afterwards an SPI controller can pull the data off SPI. Tested with ESP32:
https://github.com/luxonis/depthai/compare/spi_esp32
https://github.com/luxonis/depthai-python/compare/master...spi_esp32

The ESP32 app is based on this example: https://github.com/espressif/esp-idf/tree/v4.0.1/examples/peripherals/spi_slave/sender
with these modifications: https://github.com/luxonis/depthai-python/commit/619d485

Thoughts?

Also, to use this on existing hardware (e.g. BW1098OBC), here is the pinout information:

Thanks,
Brandon

sanjith

https://github.com/luxonis/depthai-python/commit/619d485 I went through this completely and the one which this is been inspired form. Well i could see handshake , but couldn't notice the way the model is been fed to movidious or the the way it predicts using movidious. please guid me

Brandon

Hi @sanjith ,

So SPI support is still under development. So right now the code (binary running on the Myriad X) and model are still being loaded over USB.

And then the SPI communication is done after the device is booted over USB.

So we are in the process of ordering future designs which will allow DepthAI to boot over NOR flash (and then also eMMC, but a different model). See here for example of a board with integrated ESP32 and NOR-flash boot. And then we also do have extremely limited number of NOR-flash-boot samples for interfacing with other embedded systems now here. The high price is temporary while the stock is so low... we don't want to stock out.

So later there will exist the capability for a SPI controller to load the neural network and binary code to DepthAI over SPI. We have not written this yet.