Where in the World is Roversa: What We’re Learning About How Students Engage with Robotics in Real Classrooms

We’ve been testing Roversa across classrooms in Virginia and what students are doing with it is reshaping how we think about STEM learning.

Over the past few weeks, we’ve had the chance to spend time with students across Virginia from 2nd graders in afterschool programs to middle schoolers in their CTE class to high school AP Psychology students visiting UVA.
Different ages. Different contexts. Same robot.

But this wasn’t just about sharing Roversa, it was about testing what comes next.

We brought early ideas and in-progress designs into real learning environments. These weren’t polished experiences. They were intentionally unfinished, so we could see how students interacted, where they got stuck, and what they naturally wanted to do next.

And what we learned is already shaping the next iterations of Roversa.

Personalization Starts Earlier Than We Expect
​
At a time when schools are investing heavily in AI tools that are increasingly abstract, one of the biggest challenges we’re seeing is that students are craving something they can see, control, and shape in real time. This is the space we’re designing Roversa to support.

With our digital pet prototype, it didn’t take long for students to move beyond figuring out the robot. They began naming it, talking to it, and deciding who it should be. In one session, a student paused and said she needed to “get into her creative mode” before deciding what to program.

Across ages and settings, students treated Roversa less like a tool and more like something they could shape. A button programmed to “dance” wasn’t just an action, it became a way to make the robot happy. Even students who initially seemed hesitant were drawn into experimenting once they saw how their inputs changed behavior.
​
Personalization isn’t something to add later. It’s where students begin to make learning their own.

​Feedback Matters

As students explored, they naturally started asking the kinds of questions we hope for: why didn’t it move? What happens if I try this instead? They tested, adjusted, and tried again.

But when the system didn’t respond clearly, momentum slowed.

If the robot didn’t move and there was no visible reason, students hesitated. If outputs were hard to interpret, they weren’t sure what to change. Even small delays led them to repeat actions, unsure if anything had worked.

The issue wasn’t effort or curiosity, it was feedback.

When students can see the connection between what they do and what happens next, they keep going. When they can’t, they pause. That gap shapes whether students stay engaged in the work.

The First Few Minutes Decide Everything

The biggest shift wasn’t about what students could do, it was how they started.

When we introduced more complex ideas right away, like training motion-based AI models, students hesitated. Not because they couldn’t do it, but because the purpose wasn’t clear.

But when we started with something simple and familiar, like interacting with a digital pet, everything changed. Students jumped in. They tested. They laughed. They iterated. And once they understood the idea, they were ready to go further—adding behaviors, experimenting, and making it their own.

It wasn’t about removing complexity. It was about sequencing it.

Starting with something students understand, giving them early success, and then building from there creates the conditions for deeper engagement.

​What We’re Exploring Next

One of my favorite moments came from a student who said he liked the activity because he could program the robot to do anything he wanted.

That sense of “anything” is what these early explorations are really about. Not just testing features, but understanding how to build tools that students feel they can shape, explore, and grow with.

And it’s leading us to new questions:

• How do we make feedback more visible in real time?
• What does personalization really look like in a physical computing environment?
• How early can students start shaping intelligent systems—not just using them?