A robot that explains museums — and just took 2nd place nationwide.

MuseBot AI is a self-improving museum guide robot, built by three NebuCoders members for WRO Future Innovators 2026. It earned 2nd place nationwide at the WRO Bangladesh National Final — and the team is now at a selection bootcamp, one step from representing Bangladesh at the WRO 2026 International Finals in Puerto Rico, USA.

WRO Future Innovators 2026 Theme: Robots Meet Culture Team Nebucoders — Bangladesh
MuseBot AI, fully assembled with its camera mast up and LED strip lit purple, on a table.
2nd place, WRO Bangladesh National Final 2026
3 student engineers, one robot
Bootcamp current stage — next stop, Puerto Rico
The three-person Nebucoders robotics team

Team Nebucoders

Three students — Adal Zuhair Bhuiyan, Md. Hafizullah, and Sujail Bin Abdullah — designed, wired, and coded MuseBot AI end to end. Per WRO rules, every build decision, every line of code, and every booth design on the table is the team's own work.

What MuseBot AI actually does

A museum guide robot for the small and regional museums that can't staff one — it walks a fixed route between exhibit booths, stops automatically, and talks.

Explains

Arrives at a booth, taps an RFID tag stationed there, and delivers a spoken explainer built from a human-curated fact sheet for that exact exhibit.

Listens

Asks visitors if they have questions, transcribes what it hears, and answers on the spot — referencing what the same visitor already asked at an earlier booth.

Remembers

Every question ever asked at a booth, across every tour, gets logged and folded into that booth’s explainer for the next visitor. It gets more informative the more it’s used.

MuseBot AI, fully assembled with its camera mast and LED strip, on a lace tablecloth.
MuseBot AI, chassis assembled and wired for a bench test.
MuseBot AI parked next to a laptop showing its AI conversation pipeline running.
Running the conversation pipeline against a laptop before it ever saw a museum booth.
Open chassis of MuseBot AI showing the Arduino Uno, Raspberry Pi 4B, motor driver, and battery pack wired inside.
Inside the shell — Uno, Pi, motor driver, and battery pack, wired by hand.
A hand adjusting one of MuseBot AI’s drive wheels during assembly.
Mounting a drive wheel during assembly.

The engineering journey

The real lessons are in the mistakes and the cuts, not the clean final diagram.

  1. Started big, cut to what could actually ship

    The first draft was Ubuntu, ROS2, and SLAM-based navigation — a real architecture, and a six-to-twelve-month one. Three students building for a five-minute judging slot needed a different bet.

  2. Line-following over free-roam navigation

    A fixed taped route between booths trades "goes anywhere" for "goes exactly where we told it to, every time" — the right trade when the robot has to work in front of judges on the first try.

  3. A tape stripe instead of a vision pipeline

    QR codes and camera-based object recognition were both considered and dropped. A robot on a fixed track always meets booths in the same order — a perpendicular strip of tape crossing its path is enough to mark a stop.

  4. Honest dead-reckoning over half-working steering

    Swapping the 8-sensor QTR array for two simpler FC-51 sensors meant giving up proportional steering correction entirely — a binary on/off sensor can’t produce the error signal steering needs, so the team chose a straight, honest drive over a correction scheme that would only sort-of work.

  5. A sturdier board, mid-build

    Repeated upload failures on an Arduino Nano traced back to its fragile Mini-USB port — not the code. Same chip, same pin map, swapped for an Uno’s full-size USB-B port.

  6. RFID, so exhibits aren’t locked to arrival order

    An RC522 reader lets any booth’s exhibit be rebound to a tapped card from a small web panel — exhibits can be physically rearranged between booths without touching a single line of code.

How it works

01

Two brains, one robot

An Arduino Uno owns millisecond-precise driving and stop-tape detection — no AI, no network, nothing that can jitter. A Raspberry Pi 4B owns everything smart: speech, the LLM conversation, memory. They talk over a single USB cable.

02

RFID exhibit binding

Which exhibit sits at a physical stop is bound live from a web control panel, not hardcoded — tap a card, assign it to an exhibit, and the robot resolves the right knowledgebase on arrival.

03

Free to run

Chat and speech-to-text run on free API tiers; text-to-speech runs through edge-tts at no cost at all. A full multi-booth tour costs a fraction of a cent.

Per WRO Future Innovators rule 6.5, the team's project report discloses every use of AI in the build. Claude helped implement decisions the team had already made — the wiring, the Arduino sketch, the Python conversation pipeline — while every architecture choice and every hour of physical construction stayed the team's own. It's the same "AI drafts, you verify" standard NebuCoders holds every project to.

The road to Puerto Rico

A 2nd-place finish at Bangladesh's national final put the team through to the WRO Bangladesh National Selection Bootcamp. Make it through that, and MuseBot AI represents Bangladesh at the WRO 2026 International Finals in Puerto Rico, USA.