Earlier case study
How classroom research turned speech-failure, teacher mediation, and child attention into a calmer interaction concept for a teaching-assistant robot.
I worked on the UX research and interaction design: observation synthesis, teacher-interview synthesis, design implications, robot behavior decisions, and the theatrical robot evaluation flow. The concept, report, and course outcome were team output by three designers, so this page treats the findings and final direction as shared work.
The real challenge was not to make the robot impressive. It was to make it useful in a classroom, where the teacher, the children, and the robot needed to stay aligned with the lesson instead of competing for attention.
That made the work a trust and behavior problem as much as an interface problem. The interaction had to keep lessons moving when speech failed, recover cleanly, and preserve teacher control without flattening the children's role in the exchange.
The study came from Tampere University's User Experience in Robotics course. The team worked with a language-learning robot in local preschool and elementary-school settings, controlled through a cloud-based application.
It sat inside a classroom routine with a teacher, a lesson plan, and children who needed the interaction to feel safe, readable, and worth their attention.
The primary users were children learning Finnish or another second language. They were curious, direct, and quick to test social cues. In practice, engagement depended on feedback, turn-taking, and whether the robot felt responsive enough to hold attention.
Teachers were the operators and the trust holders. They needed the robot to support a lesson without adding friction, confusion, or anything that felt unsafe or too open-ended for the classroom.
The research started with literature review and then moved into the field. The team ran two observation sessions and interviewed the teachers afterwards. Notes were captured during the sessions and then synthesized into an affinity diagram.
In both sessions, the children approached the robot with obvious enthusiasm. They were careful with it, curious about it, and eager to get a response. That enthusiasm was useful, but it also meant that the robot's behavior had to be clear enough to hold attention without becoming chaotic.
Speech recognition was the most visible failure mode. Children tried spelling commands, speaking closer to the robot, and asking the teacher to repeat the same command because they had already learned that adult speech was recognized more reliably.
Observation session in the classroom
The robot also tended to orient toward the teacher, even when the children were the ones trying to establish contact. In practice, the teacher had to turn the robot manually so its face and eyes stayed with the group.
The affinity diagram exposed a set of interaction issues that were broader than speech alone: feedback, motivation, ethics, expectations, guidance, and how the robot should acknowledge presence in a group setting.
Themes grouped from observation and interview notes
The design implication was straightforward: if the robot could not understand a child, it had to say so. Silence or vague recovery would break trust. The interaction also needed smaller, simpler turns so the robot could ask one thing at a time and avoid asking the whole group to solve a noisy failure mode at once.
Gaze was another trust signal. A robot that kept looking at the teacher was sending the wrong message in a child-facing lesson. The children needed the robot to acknowledge them directly, distribute attention more evenly, and behave in a way that felt fair in a group.
The concept aimed to make the robot calmer, more explicit, and more teachable. The team wrote a dialogue where the robot asked one question at a time, used contextual gestures to make meaning visible, and gave clearer verbal recovery when it did not understand an answer.
Immediate feedback became a core behavior decision. The "candy eyes" reward was meant to appear after correct answers, and the application gained a button the teacher could use when the robot missed a response but the child had still answered correctly. That kept the lesson moving without forcing the child to repeat endlessly.
Candy eyes as immediate positive feedback
The same logic shaped the robot's posture and gaze. It should feel less like a machine waiting for commands and more like a present participant in the lesson, while still staying visibly bounded by the teacher's control. That balance mattered because the children could treat the robot as a social actor, but the classroom still needed adult mediation.
The evaluation used the theatrical robot method. A person played the robot's role so the team could test the interaction concept without implementing the full hardware or software behavior first. That made it a practical early-stage check for the flow, but not proof of classroom impact.
Theatrical robot evaluation with teacher mediation
Teachers responded positively to the idea of the robot as a motivational assistant and pointed to several ideas worth keeping, including clearer feedback, repeat prompting, and more child-facing attention. They also reminded the team that younger children still need a teacher present and that the robot should not be framed as a replacement for classroom instruction.
Looking back, the strongest pattern in the work was research data turning into interaction decisions. The team used observations and teacher interviews to define failure modes, then translated them into behavior decisions: failure messaging, gaze, feedback timing, teacher mediation, and group fairness.
The limits are equally clear. The concept never became a field-tested classroom product, so it should be read as archive work that proves method rather than impact. Even so, it shows a pattern that still matters in my work: identify where trust is likely to break, then design behavior around that reality instead of around a polished demo.
The project did not ship as a classroom product, so the strongest public evidence is the research archive around it: observation notes, theatrical evaluation, and external coverage. Together they support the claim that the work translated field data into interaction decisions rather than novelty-driven concept art.
