Mari Velonaki is a social roboticist who has spent 25 years designing robots and interactive systems. At UNSW Sydney, she leads research focused on human-robot interaction and how autonomous systems can better support people in real-world environments.
Connect on LinkedIn ↗We tend to expect strange things from robots.
We want them to understand us, move perfectly, never get tired, never make mistakes and somehow behave more predictably than the humans who designed them.
Mari Velonaki has spent 25 years studying why that expectation misses the most interesting part of robotics.
Velonaki is a social roboticist and Director of the Creative Robotics Lab and the National Facility for Human Robot Interaction Research at UNSW Sydney. Her work asks a deceptively simple question: what actually happens in the space between people and technology?
For Velonaki, the future of robotics depends on understanding that space. It means designing for real environments, real needs and people who behave in ways researchers cannot always predict.
It may even mean accepting that a good robot does not have to appear perfect.
"It's okay to be vulnerable. It's okay if you're a little bit confused. It's okay if you don't know," she said near the end of the conversation. "Why are robots have to be so perfect when we're not, right?"
Velonaki describes social robotics as what we now increasingly call "embodied AI."
These are physical robotic systems that exist outside factories and become part of society. They interact with everyday people, not trained specialists.
That distinction matters because society is complicated.
Velonaki's Creative Robotics Lab was established around 12 years ago with a multidisciplinary approach that brought together mechatronics, software, art, design, psychology and ethics.
She is clear that those perspectives cannot be added at the end of the process as a review layer.
"It's not complimentary, it's essential," she said.
If robots are going to become part of the "societal fabric," the people designing them need to think beyond whether the technology works.
Otherwise, Velonaki warns, "we keep developing systems that they're just expensive gadgets and they're not relevant to who we are."
When asked what a truly human-centered robot looks like, Velonaki first moved away from appearance.
The more important question is what the robot does.
A system designed for a museum will have different requirements from one designed for a nursing home. A robot helping an 85-year-old remain independently at home needs to understand a different environment, set of behaviors and emotional context.
Velonaki's definition of a good robot is closely tied to agency.
"To me a good robot, it's a robot that assist us to enhance our humanity."
She wants technology that does not constantly remind people of what they cannot do.
"It doesn't remind us of our disabilities but it help us have more abilities."
That becomes particularly important as people age or their physical and emotional needs change. Velonaki argues that useful systems should learn, adapt and become personalizable over time.
The point is not to create a permanent technological replacement for existing relationships.
A robot should occupy a specific role and help people participate more fully in their own lives.
One of the most unexpected insights from Velonaki's research came while working with technologists and innovators at a large Japanese company.
When the group discussed qualities they would value in a robot sharing their workspace, one kept appearing: vulnerability.
Every system eventually fails. Robots are physical machines with batteries, materials, moving components and changing environments to deal with.
The question is what the machine does when failure happens.
Velonaki offered a simple contrast. If a robot suddenly stops functioning, a person might respond with: "Oh stupid robot, that's bad technology."
But what if the robot communicates that something is wrong before it stops?
"I'm not feeling well. I think I need to stop."
The underlying technical failure has not disappeared. The interaction has changed.
That is why Velonaki sees behavioral design as such an important part of robotics. How does a machine move? How does it sound? How does it communicate?
"It takes a village to raise a child, it takes a village to raise a robot," she said.
Healthcare makes Velonaki's philosophy particularly concrete.
She rejects the idea that robots should arrive to replace nurses and caregivers.
"We don't want to lose our nurses, right?"
Instead, robotics can take over specific tasks that make it harder for healthcare workers to provide the kind of support humans are especially good at giving.
Velonaki gave the example of lifting or transferring patients.
Even with assistive equipment, moving a person's body can create significant physical strain for nurses. It also increases their cognitive load because they need to make sure the patient is safe throughout the process.
A robot could handle the heavy lifting while the nurse stays beside the patient.
"The human is there by the side to hold your hand and talk to you, have the eye contact."
In that arrangement, technology protects the caregiver's body while giving them more capacity to provide reassurance and human connection.
Velonaki describes this as a "symbiotic and multiplayer relationship," rather than a relationship of replacement.
Velonaki also discussed something much more personal.
Her mother had Alzheimer's, and Velonaki was her sole caregiver for three years before her mother moved into a nursing home.
She understands what repeated questions can mean for both the person experiencing memory loss and the person caring for them.
Someone in the early stages of Alzheimer's may repeatedly ask when a family member is coming to visit. They may not know whether they have asked the question once, ten times or many more.
The anxiety is real. So is the exhaustion for the caregiver.
"Robots don't judge and you don't have the same social stigma," Velonaki said.
A robot can answer the same question repeatedly. It can encourage someone practicing a new skill or going through rehabilitation without becoming impatient.
"You can ask them 10,000 times."
This is where embodiment also becomes important to Velonaki. A physical system can approach someone, move, feel different materials and occupy the same space in ways a virtual agent cannot.
The technology does not need to recreate a human relationship. Its value may come from creating a different kind of support.
After decades of research, Velonaki is remarkably comfortable saying that people still surprise her.
Asked what she had learned from testing robots with the public instead of only inside controlled laboratories, she answered immediately:
"Melanie, it's a different world."
People interacting with unfamiliar technology are unpredictable.
"The hardest player is the human."
Velonaki says she is wrong about expected reactions "most of the times." Something she thinks people will dislike can fascinate them. Something designed with one intention can produce an entirely different interpretation.
Her Fish-Bird project offered plenty of examples.
The installation featured two robotic wheelchairs communicating through a love story. Visitors created their own interpretations of the robots' behavior. Children patted them to encourage them to print more messages. A composer wrote music for them. In Austria, someone even tried to steal one of the robots after interpreting its instructions as a request to be set free in a forest.
When a technical error caused the robots to spin because they could not locate one another, visitors saw something else entirely.
"Oh, look. They're like wild animals. They're so frustrated," Velonaki remembers people saying.
"We tend to connect the dots and create our own stories."
That human tendency is exactly why testing robotics in real environments matters.
Another experiment challenged Velonaki's own assumptions.
Her team studied reactions to Hiroshi Ishiguro's Geminoid, a highly humanlike robot, in both Sydney and Tokyo.
Velonaki expected people in Australia to be less accepting of such a human-looking robot.
The findings were more complicated.
Participants in Tokyo trusted the robot more. Participants in Sydney trusted it less but liked it more.
The study included people ranging approximately from 18 to 78 years old.
For Velonaki, experiences like this reinforce the need to design for specific people and contexts rather than assume there is one universally successful form of human-robot interaction.
"At the end of the day is the behavior, how something behaves is the biggest trigger for people to have some sort of emotional connection or intellectual interest."
Appearance can draw someone in. Behavior determines much more of what happens next.
Velonaki sees meaningful applications for robotics in aging, independent living, disability support, care and autonomous transportation.
But her vision of the future comes with a clear boundary.
She hopes humans and artificial systems will co-evolve and become more collaborative. Robots should remain assistive, including when decisions are involved.
"I don't believe the human in the loop. I believe the human should be on the top."
She connects that principle directly to responsibility.
"We need to be the authors of our own narratives as humans."
This becomes particularly important as AI and robotic systems collect increasingly intimate information about their users.
Velonaki raises questions about who controls the data produced by conversations about health, confidence, work, vulnerability or personal crises.
For her, responsibility cannot sit with researchers alone. Academia, industry, governments and society all have a role in defining the rules.
"We all have a saying and how we shape our future is something we have to be very vocal about."
Science fiction trained many of us to ask when robots will finally become as capable as the machines we imagined growing up.
Velonaki asks a different set of questions.
Do we need this robot? What environment will it enter? Who will interact with it? How will it communicate when something goes wrong? Will it learn as the person's needs change? What happens to the data it collects?
And perhaps most importantly, does it help people participate more fully in their own lives?
Velonaki believes the recent progress in AI makes embodied AI one of robotics' next major chapters. But smarter models alone will not solve the harder part.
"A robot that doesn't learn has no future," she said.
Neither does a robot designed without understanding the humans around it.
Near the end of the interview, Velonaki made another point that extends beyond robotics research.
"If I stay in the lab, I'm irrelevant."
Building the future requires researchers, companies and the public to remain connected. The people who will eventually live with these technologies need a voice in how they are designed.
"We all have a saying to our future."
