Picture a RoboCup@Home week: a team, a robot, five days of scored tasks. Here is what that actually looks like.

The robot stack

A customised TIAGo with our perception and dialogue modules. Key elements: YOLO-based object perception, a custom spatial memory graph, a dialogue stack built around a fine-tuned 7B model that runs on-board.

Six moments from the week

  1. Monday, 6am: calibration run before anyone else is on the floor.
  2. Tuesday: the robot picks up a plush toy the wrong way; we fix the grasp during lunch.
  3. Wednesday: the speech-to-text fails on Portuguese accents; we patch the fallback pipeline in forty minutes.
  4. Thursday: the “Help Me Carry” task. We place.
  5. Friday: “General Purpose Service Robot”. We place again.
  6. Saturday: the ranking screens — and whatever number is next to your name.

Why this matters back home

What a team builds for the competition outlives it — perception, dialogue and manipulation stacks feed straight into assistive-robotics research. Competitions aren't the point; they are a deadline.

Prospective MSc students: the robotics group is hiring students every year.