Building ADAS Modules to Make Trucking Safer at PACCAR

Autonomous vehicles fascinate me because of how many hard engineering problems they present. Most companies work on robotaxis that never have to worry about an 80,000 pound trailer swinging into adjacent lanes. But at PACCAR's Innovation Center, that's exactly what we had to solve. During my internship, I got to build a system to help prevent these massive trailers from causing accidents.

Trailer tracking comes down to precise measurement. Without constant awareness of trailer position:

• Jackknifing becomes hard to predict

• Tight turns become dangerous

• Lane changes require more guesswork

Existing solutions either cost too much or weren't accurate enough. However, we needed to make sure our ADAS software was accurately accounting for the trailer. That's where I came in. My task was to find the sweet spot between cost and precision.

After breaking down previous approaches, I noticed most went for complex custom hardware and software. Instead, I asked: "What if we used off-the-shelf sensors but got smarter about the math?"

This led to a three-part solution:

1. Standard sensors in reinforced casings

2. A mathematical model of trailer rotation dynamics

3. Integration code in Python/ROS to tie it together

The interesting part was deriving the rotation model. I can't talk about it much but using lots of linear algebra, my model was able to predict trailer position with cheaper sensors because it actually understood how trailers actually move.

Theory is one thing, but real trucks are messy:

• Sensor casings had to handle road debris

• Code needed to run in real-time

• ROS nodes had to play nice with existing systems

Each problem pushed me to learn something new - from CADing in Creo for the first time to reading books to learn how ROS works.

After several iterations, I achieved position accuracy within 2-3 degrees. I was proud of this - the code worked, my model was accurate, and it hit our cost targets. But I almost messed up the demo.

The morning of my end-of-internship presentation, I was showing my boss what I'd built. His reaction? "So... we're just going to watch numbers scroll by in a terminal?" He was right - I had spent months on complex math and system integration but hadn't thought about how to actually demonstrate it worked. We spent that afternoon building a simple simulator so people could actually visualize the trailer movements.

The experience also showed me why I love robotics: you get to connect theory (the math model) with reality (actual trucks on actual roads) and see if your assumptions hold up. Sometimes they don't, and that's when the real engineering starts.