One week. Arrive, get assigned to a team, build an autonomous rescue robot, compete. We won.
The challenge
The scenario was a simulated disaster environment: a wrecked building with victims (markers) inside. The robot had to navigate the space autonomously, find and identify the targets, and report back — without human intervention at any point. Navigation, perception, and decision-making each individually are hard problems. Making all three work together reliably, in an unknown environment, in a week, is something else.
We worked on a four-wheeled platform, using ROS to integrate the different modules. My focus was on localization and mapping — taking sensor data and building an environment model that the navigation layer could use. The integration work, making perception feed cleanly into motion planning, was where most of the debugging time went.
What the week was like
Time pressure changes how you work. You can't perfect anything; you have to decide constantly which failures matter and which don't. Which edge cases are worth handling? Which parts of the system need to be bulletproof and which can be fragile? A week is enough to build something that works, but not enough to make it clean — and the distance between those two things is instructive.
Working alongside engineers from ETH and other strong programs also recalibrated my sense of what a team can do under real constraints. The atmosphere was collaborative despite being a competition, which I didn't expect but appreciated. Everyone was there to build something real.