AI Trends

Enthusiast uses Claude and Arduino to get a 1994 Cherokee to pass California emissions testing

English text is being prepared.Updated Nov 14 22:08

AI summary

American AI enthusiast Mohammad Zulali used Claude and Codex to get a 1994 Jeep Cherokee to pass California's emissions test. He used an approximately $3 Arduino Nano development board connected to the Chrysler SCI serial communication bus to read engine data, with Claude handling diagnosis and Codex handling coding; because the original vehicle computer does not support firmware flashing, he ultimately wrote custom code onto an EEPROM chip to run. The related project is open-sourced on GitHub under the name xj-scanner and uses the GPLv3 license.

Why read this

AI score 63

A complete hands-on case of using Claude for diagnosis and Codex for coding to fix an old car's emissions, demonstrating a path to solving concrete engineering problems with low-cost hardware plus AI.

Topics

Tags

Body · AI translation

With Claude and an Arduino board, an enthusiast got a 1994 Cherokee to pass California emissions testing

English text is being prepared.

English text is being prepared.

The engine computer of this old car has very limited functionality, but after Zuraili connected it to a laptop using an Arduino Nano development board, he was not only able to read vehicle data, but also gradually achieved control over some parameters.

The video he released documented the entire process of Zuraili solving the exhaust emissions problem: from repeatedly failing tests, to successfully connecting to the engine computer and reading data, to adjusting the engine and resolving the excessive emissions problem. Throughout the process, Claude was responsible for guidance, while Codex was responsible for writing code.

The Cherokee had previously failed California's exhaust emissions test multiple times because nitrogen oxide emissions exceeded the limit at specific vehicle speeds. Initially, the vehicle failed the test at 40 km/h. Zuraili then adjusted the oxygen ratio in the air-fuel mixture, but unexpectedly the situation worsened, and it failed both the 40 km/h and 80 km/h tests.

English text is being prepared.

Fortunately, in 2026 there is AI to help, and besides AI, only a 3-dollar microcontroller (at the current exchange rate, about 20.1 yuan) was needed, with all electronic components totaling no more than 15 dollars (at the current exchange rate, about 100.3 yuan).

The first thing to solve was how to make the 32-year-old engine computer communicate with a modern laptop. Claude determined that the Cherokee uses the Chrysler SCI serial communication bus, with a communication rate of about 7?800 baud, and the electrical signal polarity is opposite to that of an ordinary serial port. The Arduino Nano development board can handle the signal inversion problem, allowing the laptop to read vehicle data. Zuraili connected the development board to a breadboard, then connected several wires, three of which went into the Cherokee's 6-pin diagnostic connector.

The first connection was successful, but initially the system could only record coolant temperature and battery voltage and plot their curves. Zuraili wanted to obtain more engine data and also wanted to directly adjust some parameters to solve the excessive emissions problem.

Because he mistakenly referenced data from other vehicle models, the system twice incorrectly determined that the engine air-fuel mixture was too rich, believing that the fuel injection amount was 22% higher. After that, Zuraili had Claude collect data during test drives, recording the engine parameters and fault codes when the vehicle was traveling at 40 km/h and 80 km/h respectively, to find the cause of the problem.

Claude was also given permission to control parameters such as fuel pressure. After obtaining seemingly reliable diagnostic results and solutions, Zuraili originally intended to flash the AI-modified program into the engine computer so that the vehicle would pass all of California's exhaust emissions test items. But the Cherokee's engine computer was too old, several months earlier than the version that could be reflashed with firmware, and simply did not support this operation.

So Zuraili instead had Codex write custom code into a nearby EEPROM memory chip and run it directly. Zuraili admitted: "Claude did most of the diagnostic work, and Codex handled almost all the memory processing work. I was only responsible for wiring and driving."

The whole process also did not require hiring a repair technician, let alone relying on the special diagnostic equipment of a Chrysler dealership.

Zuraili also uploaded the project used to diagnose and solve the Cherokee's emissions problem to GitHub, naming it xj-scanner. The project uses the GPLv3 open-source license, and anyone can use it for free.

Source

English text is being prepared.