Duburi

Problem

RoboSub is a competition where an autonomous underwater vehicle has to complete a mission in a pool — pass through a gate, follow a path, find and interact with objects, and surface in the right place. The vehicle has to do all of it on its own, underwater, in real time. Duburi is BRACU’s entry: an AUV built to perform in water that doesn’t care how elegant the code looks.

Difficulty

Water removes most of the tools a robot normally leans on. Radio dies within a meter or two, so there is no reliable remote control and no GPS. Cameras look through murk and refraction. Everything drifts — the vehicle, the water, the mission objects. The environment does not care how elegant the code looks; it only cares whether the machine actually does the task.

Built

The Duburi AUV line, currently at Duburi 4.2 — a full RoboSub vehicle built by a student team at BRACU. Perception, control, mission planning, and the hull that has to survive being handled by students and water at the same time. The public home is bracu-duburi.com; the code lives at github.com/fh1m/Duburi. The substantive code lives in the duburi_ws and duburi-codebase_RND workspaces — the Duburi repo itself is a 1-commit placeholder.

Owned

I came in as a Junior AI/Machine Vision member, then led the AI & Machine Vision subteam, then became Engineering team lead. The perception stack was mine to own: making the vehicle see well enough to act, and making that vision reliable enough to trust in a mission.

Failed

The results are the evidence, not the identity: 2023 world runners-up (2nd), 2025 9th, 2026 8th (6230 points). The 2025 drop is the honest data point — the machine did not perform the way the team expected. The failure was real and it was useful: it said something about the vehicle, not about the people.

Changed

The vehicle changed across generations — Duburi 4.2 is not the machine that placed 2nd in 2023. The engineering became increasingly serious as the path became non-linear. Each competition changed what the team believed about what actually matters underwater.

Machine-now

At RoboSub 2026 the machine scored 6230 points and placed 8th. It performs in an environment that does not care how elegant the code looks — and it keeps being rebuilt to perform better next time.

Lesson

Machines that have to perform in an environment that does not care how elegant the code looks teach you to care about the environment first. Elegance is a means; the mission is the judge.