AutoProber: A homemade environment that uses a collection of parts and an AI agent to photograph and map circuit boards for probing.
gainsec/autoprober: hardware hacker's flying probe automation stack
https://github.com/gainsec/autoprober
Probing is the process of examining voltages and signals by placing needle-shaped components called 'probes' on pads and pins on a circuit board. AutoProber is a custom environment for automating this probing process using an AI agent. It identifies where to examine on the circuit board, displays potential locations on a dashboard, and proceeds to the actual probing only after the operator approves.
According to J-GainSec, AutoProber begins by having an AI agent load the project, connect all the hardware to be used, and verify that each part is functioning correctly.
Next, AutoProber performs a home operation to return the CNC machine to its reference position and performs calibration to correct any misalignment. After that, the operator attaches a dedicated header that integrates the probe and microscope.
When the AI agent is informed that 'a new circuit board or object to be measured has been placed on the stage,' the AI agent first searches for the object's location on the stage. Once the object is found, it takes pictures of it little by little, recording the position of each image along the X, Y, and Z axes, and identifies pads, pins, chips, and other notable parts on the circuit board.
The notable components that the AI agent finds include various headers and I/O banks,
The AI agent then stitches together the captured images to create a single map that shows the entire circuit board, and the locations of any pins or notable parts found are recorded on that map.
Potential probe locations found by the AI agent are added to the web dashboard, allowing the operator to review the candidates and decide whether to approve or reject them.
Approved targets are actually probed, and the results are displayed. J-GainSec explains, 'The connected hardware can be controlled from a web dashboard, a Python script, or the AI agent itself.'
The following are the parts that J-GainSec used in the AutoProber prototype.
Optical end stop
USB microscope
SainSmart Genmitsu 3018-PROVer V2
- A Matter-compatible smart power strip with individually controllable AC outlets, two USB-A ports, and two USB-C ports.
• Siglent oscilloscope 'SDS1104X-E'
• Jumper wires
- A pen spring or a similar light compression spring
3D printer for outputting toolhead parts
Additionally, the following are listed as optional or interchangeable parts:
• General oscilloscope probes
・ 5V USB power adapter
USB 2.0 splitter cable
According to J-GainSec, AutoProber needs to be treated not as a regular web application, but as a 'system where actual machinery is in operation,' and therefore, detailed safety measures are in place. For example, instead of simply trusting
Furthermore, while the machine is operating, AutoProber constantly monitors the signals entering 'Channel 4,' and stops operation if a trigger is generated in 'Channel 4,' if the voltage is unclear, if the CNC machine tool emits an alarm indicating an abnormality, or if a signal is received indicating that the X, Y, or Z axes of the CNC machine tool have reached their limit. After that, the voltage, status, and operation are recorded, and it waits until the operator explicitly releases the stop state. 'It will not automatically restart operation after stopping,' J-GainSec explained.
Furthermore, J-GainSec has made the Python control code, CAD files, and documentation for AutoProber publicly available on GitHub, so you can also assemble and try it yourself.
AutoProber has also been featured on the social news site Hacker News , where there have been positive comments such as 'A single repository that combines off-the-shelf hardware and lets an AI agent process the workflow is valuable,' but there have also been questions about the role of the AI, such as 'What is the AI doing? Does it just find all the pins, or does it do more than that?', as well as concerns about accuracy, such as 'Physical hardware operates with high precision, but AI operates probabilistically, so even a 0.1mm mistake in pin position could damage the circuit board,' and comments questioning the necessity of it in the first place, such as 'Circuit board testing has been established for decades, and if you have design data , you can determine the position of the target with high precision without using AI.'
Related Posts:
