Introduction
The robot stopped before a threshold as if the floor had suddenly formed an opinion. Yuna Toma stood in the lab with a notebook in her hand and watched an expensive system fail at an everyday edge. Nobody cheered. That was exactly why she found the moment useful.
Yuna lives in Zurich and works in a project on assistive robotics. She came from mechanical engineering and for a long time thought more in motors, joints and control loops than in neural networks.
Story of the Path into AI
In her first lab placement Yuna learned that machine intelligence has a body, even if many people talk about it as if it lived only in data. A gripper recognized a cup reliably until light from the window hit the glaze. Then it missed. The model had not had a bad day; the environment was simply less tidy than the training images.
Yuna had to learn a new technical language and everyday language at the same time. In meetings she hesitated too long before asking questions because she did not want to stand out as an exchange student who had missed the thread. Eventually she began an error notebook: every wrong movement, every failed grip, every irritated test face. Her first project was a gripper that recognized objects and displayed uncertainty when material was unfamiliar. The first version was too polite: it reported uncertainty so late that the arm was already halfway moving.
Current Work
Today Yuna tests robots in care-adjacent environments. With trays, water glasses and walkers, it quickly became clear that speed is overrated. Older test participants wanted predictable movement and clear pauses. Yuna therefore changed the model and the choreography: slower start, visible stops, simple signals.
The project remains a prototype. Yuna dislikes speaking of relief when it is still unclear what extra work maintenance, supervision and training will create. A robot can help, but a bad robot produces new worries on wheels.
Personal Advice
“An elegant failure is more valuable than a smooth demo,” Yuna says. She advises young developers not to hide test protocols. The ugly edge cases show whether a system is being built for people or just for presentation.
Key Facts
Age and place: 25, Zurich.
Background: mechanical engineering, international study path, lab practice.
Entry into AI: gripper with visible uncertainty for light, form and material.
Focus today: assistive robotics.
Typical tools: robotics, computer vision, uncertainty measurement.
Werkstattnotiz
Yuna’s error notebook has a page of its own for thresholds. It contains no big formulas, only sketches, height measurements and short sentences from test participants. Once someone wrote, “It hesitates like a dog.” Yuna still does not know whether that was praise or warning. She keeps testing slower movements.