ECLAIR Makeathon 2025

Overview

I was Lead Programmer for our ECLAIR Makeathon entry, a line-following robot built and tuned inside a 48-hour window. Ours was the only robot to complete the full course, and the team took first place.

The problem

A 48-hour deadline changes which engineering decisions are correct. There is no time to build a system and then tune it; the design has to be one that can be tuned while it is being built.

The course demanded both speed and precision, which pull against each other. A robot that follows the line conservatively finishes late or not at all. A robot that runs fast overshoots corners and leaves the line entirely.

Approach

I used PID control over multiple IR sensors. The sensors report where the line sits relative to the robot’s center; the controller turns that error into a steering correction that scales with how far off the robot is, plus terms that account for how the error is changing and how long it has persisted.

PID was the right choice specifically because of the deadline. Its three gains map to behaviors a person can observe directly: one determines how hard the robot reacts, one damps the oscillation that reaction causes, one corrects a persistent drift. That made tuning an observational loop rather than a code-editing loop, which is what fit the time available.

Challenges and tradeoffs

Multiple IR sensors give a finer sense of position than one, but they only help once you know what each reports on the actual course surface. Reflectivity differs between materials, so the thresholds had to be set against the real course rather than assumed.

The speed-versus-precision tension resolved into the gains themselves. Too aggressive and the robot oscillated down straightaways; too gentle and it left the line at corners. Finding the balance was iterative and could only be done on the real course.

Result

The robot completed the full course, the only entry that did, and the team placed first. The source is public.

What I learned

Under a hard deadline, the best design is the one with the shortest feedback loop. PID won here not because it was the most sophisticated option, but because its parameters corresponded to things I could watch the robot do and fix in the next run.