AI CAN bus reverse engineering |
|
|
|
|---|
|
CSS Electronics newsletter - July 2026 |
|
|
|
|---|
|
|
AI CAN bus reverse engineering [Claude Code] Can an AI agent reverse engineer your proprietary CAN bus signals?
Martin Falch tests this using our CANsub interface, python-can and a custom Claude Code 'skill' to decode speed, RPM and more.
Quick context: Reverse engineering proprietary CAN signals is an extremely common use case (e.g. for mixed fleet telematics - or just for fun). However, doing it by hand is slow and demands deep expertise. This workflow lets you do it with minimal expertise - and in 95%+ less time vs. the 'classic' approach.
The intro covers the below topics: |
|
|---|
|
| | The skill
We explain how the Claude Code skill works (instructions + context material + Python scripts) and how Claude uses it to dynamically reverse engineer proprietary CAN data. | | |
|
|
|
|---|
|
| | Reference signals
Learn why 'reference signals' are vital. This can be e.g. a CAN-based signal (like OBD2, GPS-to-CAN, ...), a filmed dashboard value (leveraging vision OCR) - or noisy human input via a built-in browser app. | | |
|
|
|
|---|
|
| | Step-by-step analysis
Get the technical breakdown of what happens under-the-hood incl. how Claude performs bit-activity surveys, correlation ranking, exhaustive bit searching for start bit/length/byte order/sign and finally scale & offset fitting with physical sanity checks - and how the result is formalized in ready-to-use DBC files - all using python-can. | | |
|
|
|
|---|
|
| | Why use the CANsub?
4 key reasons why we recommend using the CANsub.2 as your CAN sniffer interface. | | |
|
|
|
|---|
|
| | How to get started
We provide step-by-step guidance to enable you to try this yourself within 10 minutes - incl. free sample data and the 100% open source skill. | | |
|
|
|
|---|
|
| | Ex 1: Mercedes speed & RPM (OBD2)
Using OBD2 PIDs as reference, Claude decodes proprietary speed + RPM signals in 5 min - producing a ready-to-use Mercedes DBC file. | | |
|
|
|
|---|
|
| | Ex 2: Opel Astra speed (vision OCR)
If no CAN-based reference is available, we can sometimes fall back to vision-based OCR (Optical Character Recognition). Here, the car's dashboard speed is filmed and digitized - letting Claude decode the proprietary speed signal. | | |
|
|
|
|---|
|
| | Ex 3: Gauge position (human input)
Claude single-shot decodes a gauge position signal from a noisy 'human input' reference signal (via a built-in browser app) - and extrapolates it to 7 other gauges. We then verify the validity live via webCAN. | | |
|
|
|
|---|
|
As the intro shows, this concept works extremely well. We believe this will enable a real democratization of CAN sniffing - and a potential 95%+ time reduction per signal.
Tip: The intro is also available as a 30 min YouTube video at the top of the article.
Learn more below - and try it yourself with our free sample data pack!
|
|
|---|
|
|
CSS Electronics
Soeren Frichs Vej 38K
8230 Aabyhoej
+45 91 25 25 63
contact@csselectronics.com |
|
| |
|---|
|
This email was sent to dupa@niepodam.pl |
| | |
|
|
|---|
|
|
|
