Could a robot fix the loneliness epidemic?

Note: Transcripts are automatically generated from episode audio, and are not fully corrected for spelling, grammar, and formatting.

GRACE BROWN: Abby is a social humanoid robot. She’s about 110 centimeters tall, the size of a young child, very bright and colorful, and she has a lot of personality.

BROWN: She’s very inquisitive, very curious, and very childlike in nature. When people think of humanoid robots, they think of the Terminator. That is not Abby at all. People just lighten up a lot and realize that robotics doesn’t have to be something that’s aggressively masculine and intimidating.

RANA EL KALIOUBY: What are you most excited about this year?

BROWN: I want to walk down the street one day and see more Abbys in the streets than people walking their dogs. That’s what I’m excited about.

EL KALIOUBY: Wow.

Tell us where you are in your journey.

BROWN: Everyone talks about the robotics industry like it’s coming, it’s going to happen in the next few years, it’s around the corner, et cetera. But for us, it’s not coming. It’s already here.

EL KALIOUBY: On today’s episode of Pioneers of AI, I’m interviewing Grace Brown. She is the co-founder and CEO of Andromeda. They are a company building a social humanoid robot designed for the aging population. This isn’t your typical Terminator-looking social robot. This one is more pint-sized, and it even blows bubbles. So let’s dig into it.

I’m Rana el Kaliouby, and this is Pioneers of AI, a podcast taking you behind the scenes of the AI revolution.

Hi, Grace. Welcome to Pioneers of AI. It is so great to have you on the show.

BROWN: Thanks, Rana. I’m super excited to be here.

Copy LinkHow a teenage engineering class sparked a robotics career

EL KALIOUBY: You’ve been building robots since you were 15, which is so cool. How did you get started?

BROWN: I was very fortunate. I went to a really good high school, and we had an engineering subject as part of Year 10 and Year 11. That was really my introduction to robotics and engineering as a whole discipline. Prior to that, I’d always, I think like most engineers, thoroughly enjoyed math and physics.

I did quite well in school in those subjects. I found them to be a lot of fun. When I was really young, I wanted to grow up and be a mathematician. Then when I took that first engineering class, we had to build this three-degree-of-freedom hydraulic arm, and it was just so much fun. That’s kind of what got me started. It also introduced me to Arduinos and MATLAB and writing code. So that’s where it all started.

EL KALIOUBY: Fun. I love Arduinos. When my kids were younger, we would all do Arduino projects together. So then you went on to start your company, Andromeda. Andromeda builds social robots, and your robot is called Abby. Tell us about the origin story of the company. Why did you start it?

BROWN: I was doing my mechatronics engineering degree at the time.

I had my degree during the COVID-19 pandemic, so a lot of my classes were online. I was living in a student dorm at the time. It was during that period that the idea to build Abby really started.

EL KALIOUBY: So you wanted to build a robot friend or a robot companion. What was the impetus?

BROWN: At the time, I had really modest ambitions. I wanted to build myself a system that could ask me how my day was going and give me a hug, because I’d realized that for about four or five months, I hadn’t had a hug from a friend or family member. It started off with very modest goals, as well as partly wanting to exercise my muscles as a robotics engineer and work on a project that could really upskill me and push me technically. So it was a combination of all of those things.

Copy LinkTurning pandemic loneliness into a mission for elder care

EL KALIOUBY: When did you realize there was a real business opportunity here and that you needed to officially found a company?

BROWN: It was when I was coming out of the pandemic and thinking about the experiences I had of feeling really isolated and disconnected from the world. The pandemic really shined a light on the loneliness epidemic and the health implications of it as well.

The health effects are equivalent to smoking 15 cigarettes a day. But the people most affected by this are our older generations, older adults, and seniors. When I started looking into the market, I realized that many people living in aged care homes today experience what I experienced during COVID-19 on a regular basis, not just during a lockdown or a pandemic.

So I started going to local aged care homes. I brought Abby in in a basket at the time, just to see if she could really add some value for this industry.

EL KALIOUBY: What are some of the ways Abby is being used today?

BROWN: Every individual uses Abby slightly differently. But the way I’d break it down is that there are really two areas of interaction: group interactions and one-on-ones. In the group activities, Abby does what I’d call more performative activities. She does dances and exercises, meditations, quizzes, all the activities that get people a bit more socially engaged in a home.

In the one-on-one interactions, it’s very much about building a relationship with individuals. That means having regular check-ins with them, asking them how their day is going, and speaking with residents in their native language, which is so important for this demographic because, depending on your cognitive decline, if English is your second language, you’ll often revert back to your native language. So communication is a massive barrier in the industry.

What we found is that Abby being able to go into a home and just check in on a resident and ask them how their day is going in their native language has dramatically increased the quality of their life.

Copy LinkWhy Abby is designed to feel warm and approachable

EL KALIOUBY: I want you to bring Abby to life for us, because you were very intentional about your design decisions and how you built Abby. What does Abby look like?

BROWN: We’ve modeled her off the size and physique of a young child. She’s about 110 centimeters tall. I think that’s about 4 feet tall. She’s bright and colorful. She’s anthropomorphic, so she’s got arms and legs that mimic human movements.

She’s got these bright, large, oversized purple eyes that kind of stare at you. All of her designs were meant to be very humanlike in terms of being relatable in the movements and gestures, but not so human that it crosses the uncanny valley curve.

EL KALIOUBY: Explain the uncanny valley for folks who are not familiar with it.

BROWN: The uncanny valley is basically when something tries to mimic being very humanlike, but you can tell something’s off. So it makes you, or makes the audience, a little uncomfortable, a little creepy.

Something is just a little bit off. Lots of people who are building humanoid robots with fake prosthetic skin or faces often make people really uncomfortable because they can identify that it’s not a real person.

So we wanted to make Abby look very robot-like, but also still make her feel real. Real in the sense that when you interact with her, you feel like she’s this brought-to-life character that you’re having a real interaction with.

When you look around the market, a lot of other humanoid robots, even if they’re not crossing the uncanny valley curve, are copying science fiction really closely, building these really tall, militant Terminator-like robots. We didn’t want to go down that path either, because if we’re going to go into people’s homes, people have to feel comfortable having this product and this system in their homes.

People have to have a lot of trust in what they’re adopting. So that’s also what we were optimizing for: something that was very approachable, nonthreatening, and quite endearing.

EL KALIOUBY: I want to double-click on that because it was so refreshing to see a social robot that is warm. There’s a lot of funding going into the physical AI space in general, but specifically humanoid robots.

But I look at some of them and think, wow, whoever designed these things, it’s so hard to imagine this thing roaming around my kitchen, even if it’s the most helpful robot ever. I think a lot of these robots are missing the human element and the social and emotional capabilities.

BROWN: Exactly. When I was thinking about Abby, all of my inspiration actually came from Disney and Pixar characters from childhood. People have always connected with these characters on such a deep level, and we know that they’re fictional.

We know that they’re not actually real, but they still invoke these very real emotions from their viewers and the audience. Different people see different parts of themselves in all these characters, and that connection we have with fictional characters is exactly the connection I wanted people to have with Abby.

So a lot of what I’m trying to build through Abby and through Andromeda is actually what animators have been doing over the last century: creating story lines with arcs and narratives and characters with such vibrant personalities, and bringing that to life through robotics. That’s been my mission from day one.

Copy LinkBuilding a robot personality that adapts to every setting

EL KALIOUBY: That’s very cool. So what is Abby’s personality like?

BROWN: I recognized very early on that there isn’t a one-size-fits-all approach. She adapts to the audiences that she interacts with.

Out of the box, Abby is quite inquisitive, quite curious, very childlike, and very playful. But she does tailor and adapt her personality based on who she’s interacting with, because there isn’t a one-size-fits-all approach.

Some people want to talk deeply about philosophy or history, and other people just want to do more playful things, like dance or do tai chi, or they might want to speak in different languages. We even have a version of Abby that’s in a nursing home, and it’s a Catholic nursing home. So that version of Abby is actually a Christian Abby. She adapts to the different environments she’s in and who she’s interacting with.

EL KALIOUBY: There’s a version of Abby, I believe, that has a bubble — her hand is like a — tell us more about that.

BROWN: Yes. Her left hand is actually a bubble gun. I quite like it. This was when we had her in a children’s hospital, so we thought bubbles would be something nice and playful.

To me, what Abby stands for is contradicting the industry. When people think humanoid robots, they think the Terminator. They think of something quite intimidating. That is not Abby at all.

When they see her bubble gun, they think the same thing. They’re like, what does she have on her left hand? Then she blows bubbles, and people just lighten up a lot and realize that robotics doesn’t have to be something aggressively masculine and intimidating.

It can be something a lot more playful and enjoyable, and it can really improve quality of life for people and enrich people’s lives. That’s what I’ve wanted to work on since day one.

Copy LinkWhy the robots we build reflect the people behind them

EL KALIOUBY: There’s not a lot of women in the robotics-slash-physical-AI space. Why do you think it’s important to have more women and more diversity in general designing these robots?

BROWN: If you look at all of the robots in market today, I really think they’re a reflection of the teams that build them.

EL KALIOUBY: Say that again, Grace. Say that again.

BROWN: It’s true. I get that feedback all the time. People compare me to Abby a lot, and they always say they see my personality in her.

They also think she has my voice, but she’s got an American accent, so I don’t. But if you look even now, I’ve got lots of engineers working on Abby, and we still get comments where people who know my engineers look at Abby and go, oh, I can see Tom’s personality coming out of Abby, or this is Erica’s personality coming out of Abby. They start seeing the developers in Abby.

So it comes down to the software, but also the hardware and the design as well. All of it is just a reflection and an embodiment of the values and personality of the developers and the teams building these systems.

EL KALIOUBY: Unlike a lot of the robots out there, Abby has social and emotional intelligence. How Grace has trained her robots after a short break.

Welcome back to Pioneers of AI. You can also watch this episode and see what Abby looks like by heading over to our YouTube channel.

Copy LinkTeaching Abby to sense emotion and respond like a human

Abby obviously has eyes. That’s the computer vision part, and I’m guessing she has ears, so she’s listening, right?

BROWN: Auditory.

EL KALIOUBY: Auditory, exactly. What other sensors does Abby have?

BROWN: She’s got three main inputs at the moment. She’s got her microphones, her visual sensors, her cameras, and then she’s also got a lidar for navigation as well, although in later generations we may move to a fully vision-based navigational stack.

She’s got this microphone array so that when you speak to her, she understands where the voice is actually coming from. Voice detection as well — who’s speaking to her. Diarization in group settings is actually a really complicated challenge.

There are so many things, and building something like Abby is so complicated. Part of the enjoyment of actually building Abby is solving all of these deeply hard technical challenges.

But everything, when it comes to building Abby — from the design to the hardware to the software — boils down to the fundamental questions of what it means to be human. How do you coordinate your arms and legs and head in such a way that you’re expressing a very specific emotion or communicating very specific things?

Even when it comes to a conversation, it’s never just a back-and-forth like a very old-school chatbot. There are so many nuances in understanding who’s speaking, when they’re pausing versus when they’re waiting for you to say something.

There are a lot of nonverbal cues in communication as well. If you said something to me like, “Oh, look over there,” that’s really complicated for a robotic system to break down and understand — look at where they’re pointing, et cetera. So there’s a lot, but it is a lot of fun.

EL KALIOUBY: I spent 25 years of my career building emotional intelligence — emotion AI — into machines. One question I have for you is, Abby does have vision, but can she see facial expressions?

Can she understand if I look excited or scared or sad or mad? And can she hear it in the voice as well?

BROWN: Yes. Those are all the things we are aiming to be world-class at. I put that all under the bracket of the psychology of interaction. She’s looking at all of your microgestures and microexpressions. She’s looking at the warmth and tonality of your voice.

That’s her input. Then when she expresses things, her output, we’re doing the same as well. What is her voice prosody, the pitch, and sometimes just the silence — when to gesture versus when to sit a bit still. All of that matters too.

I think right now the industry has definitely overindexed on functionality and capability, and that’s very important.

It’s so important, and that’s going to revolutionize so many different industries. But when you think about going into a domestic environment and people’s homes, it’s going to be so important to understand the nuances of exactly how the other person is responding and reacting, and their emotional state.

EL KALIOUBY: It’s actually also pretty complex, because I remember from the work we were doing with some of the roboticists, even deciding when to mirror a person’s emotion versus try to nudge the person out of an emotional state. That’s all right — when to pause, when to respond. These are all dynamics that are kind of hard to incorporate.

BROWN: It’s so complicated. The things that I’m most excited about are around the personalization of these home companion robots.

It’s not just about understanding the facts of the individual. If you interact with Abby, she might learn your name, that you have two or three children, that they go to school in this suburb or area, and these are all the facts. But what’s going to be so powerful and actually magical is when they start understanding the nuances.

For example, if Abby’s in a nursing home and she talks to someone called Margaret, she learns that Margaret prefers to be called Maggie, and that when Abby talks too fast, that’s overwhelming for Maggie, even if Maggie never explicitly said that.

It’s going beyond just learning the facts of an individual and starting to really understand them as a person.

EL KALIOUBY: At the end of the day, over 90 percent of how humans communicate is nonverbal, right? So if these robots are going to be in our homes, interacting with us on a day-to-day basis, they need to incorporate these nonverbal signals into their input channels, which is exactly what you’re doing, and I love that.

Copy LinkHow memory personality and technical systems work together

EL KALIOUBY: What is the underlying large language model or world model that Abby uses?

BROWN: She uses quite a number of different models that are available online today. She’s got, like we talked about earlier, all of her sensory input, from her microphone array to her cameras to her lidar, and all of that is fed into her behavior tree, which then uses different models and different parts of her architecture to give a response, whether it’s a behavior or a verbal response or a gesture or suggesting an activity.

I think a lot of people have the misconception, and understandably so, that companion robots today are almost just LLM wrappers. But when it comes to building a social robotic system or a social interface, like we talked about earlier, there are so many nuances being brought in — the voice prosody, the microexpressions, the sentiment analysis, and then also structuring Abby’s personality as well.

Hardware is inevitably going to reach a failure point. Failure points are going to happen. So for me it’s also about how Abby responds to failure points.

Will it actually recognize that it’s made a mistake and apologize for it? For example, a very simple thing we do is, if the internet connectivity is really low — a very common technical challenge for robotics — Abby will actually respond and say things like, “I’m really sleepy today. Please don’t ask me too many questions.”

That’s not because she’s actually sleepy.

People find that really funny, and they’re also more forgiving of mistakes. Usually when the internet is slow, there’s higher latency and slower responses, and people are more forgiving when you have that social operating system layer and can actually apologize for the mistakes or the technical hindrances that have occurred.

EL KALIOUBY: Does that mean that all of the processing Abby does — all of the compute — happens in the cloud, or does she do any of the processing locally?

BROWN: We have a hybrid where some is in the cloud and some is local. In the environments we go into, connectivity can sometimes be unstable. So for us, we’ve actually got a hybrid where, if the connectivity is low, she uses more local models, which are obviously less engaging and less interactive.

Another thing we do as well is, when you interact with Abby, we get her to say things like “ums” and “ahs,” just like we do as humans, and there’s really no reason for a robot to do that. But first, it makes it more natural. Second, it also kind of masks the latency that she has.

EL KALIOUBY: Does Abby have memory?

BROWN: Yes, she does.

Abby has memories of who she’s talking to, all the basic facts she’s collected from their interactions, but also that deeper-level understanding. Understanding that maybe Maggie goes for a very specific football team, and that Maggie doesn’t like people interrupting her in the morning before breakfast but is very chatty after lunchtime. So Abby might not go in before breakfast, but will always make sure she does a check-in after lunch.

I think memory is going to be one of the next really major developments. I don’t know if breakthrough is the right word, but it’s definitely going to be something very significant for the field in this next wave of robotics.

EL KALIOUBY: I actually think of memory as a potential competitive moat if you can crack how to represent that in the right way. But again, it’s one of these things that is quite complex, I believe.

BROWN: It definitely is. You’ve also got to consider privacy and compliance and how you’re managing the data. So it’s not a small task, that’s for sure.

EL KALIOUBY: And it’s true, there are some real security concerns when it comes to these kinds of systems. But it’s not just about the data — it’s also about getting the physicality right. Have you seen these viral videos of humanoid robots toppling over? Honestly, they’re kind of comical, but they’re also quite concerning. We’ll get to that and more after a short break.

Copy LinkDesigning for safety trust and responsible care

EL KALIOUBY: I want to talk about safety a little bit, and accessibility, right? Especially since you’re deploying Abby with elderly individuals. What are some of the considerations you’ve had to think about? For example, what if Abby runs out of battery or is about to run out of battery?

What does she do or what does she say?

BROWN: There are major ethical and philosophical questions around this space that my team has to consider. When it comes to physical safety, that’s front of mind for us when we’re designing her. Right now, even though she’s anthropomorphic, so she’s got two arms, two legs, and her head, she’s actually on a wheeled base. So when she powers down, if the power was cut out for whatever reason, she actually doesn’t fall over. Her arms and her legs might go down, but she doesn’t actually topple over.

EL KALIOUBY: One of my concerns around the use of AI chatbots for now is around safety and guardrails around what kind of recommendations this chatbot or AI system can give you, especially around mental health. How does Abby handle a situation where a person is potentially considering harming themselves?

BROWN: We work really closely with all of our customers and partners in how we design Abby, Abby’s guardrails, how she manages data, how conversations are stored, and exactly what happens next.

Usually when conversations reach fairly sensitive topics, Abby will always default to making recommendations to speak to one of the health care workers or the home.

EL KALIOUBY: Even the Catholic Abby we had — the Christian Abby that we’ve deployed — my team tested the boundaries of the Christian Abby quite rigorously. They actually had the time of their lives that afternoon, just seeing how far it would go.

BROWN: Even though she was Christian, she was still fairly diplomatic in her responses, and she would never go too far in one direction. That’s something we consider quite seriously: making sure that although she’s personalized to the individual, if you do go into areas of conversation that are a bit more sensitive, she’s always encouraging a human to be brought into the conversation.

EL KALIOUBY: That’s great.

Copy LinkScaling Abby from Australia to a much bigger vision

EL KALIOUBY: So Andromeda got started in Australia. You raised an initial seed round. You’re deploying Abby in a number of facilities. But how has the fundraising journey been so far? What are the next milestones? Tell us where you are in your journey.

BROWN: I’m obviously Australian, that’s why I have a funny accent. So we started in Australia. We raised our last round from Fourrunner Ventures, which allowed us to start setting our roots up here in San Francisco.

We had some really phenomenal commercial traction back home. Everyone talks about the robotics industry like it’s coming, it’s going to happen in the next few years, it’s around the corner, et cetera. But for us, it’s not coming. It’s already here.

We have Abby in so many different homes where she’s interacting with people every single day. We’re hearing all these stories about the impact she’s having. So we started in Australia, then raised from some U.S. investors. I’ve just moved over to San Francisco in September 2025.

And we’ll be looking to expand here over the next few years. The nursing home industry is very similar to what we have back home in Australia. We already have some pretty strong interest from customers here. So it’s really just about expanding.

EL KALIOUBY: That’s very cool.

What are you most excited about this year?

BROWN: I’m going to give a cliché answer, but honestly, it’s just having Abby reach more people. What brings me so much fulfillment, and my team as well, is seeing the impact she has in the homes and the stories and relationships that people build with Abby.

A lot of the reason my engineers join Andromeda is for the technical challenge of building something like Abby. They’re really drawn in by that. But what really keeps our team together and motivated is seeing the impact she has in the homes. That’s so fulfilling. So having that on a much, much larger scale is what I’m most excited about.

I’m most excited in the next few years to see tens of thousands, eventually millions, of Abbys out in the world. I want to walk down the street one day and see more Abbys in the streets than people walking their dogs. That’s what I’m excited about.

Copy LinkWhat it takes to flourish as AI changes how we build

EL KALIOUBY: Wow. That’s a big vision. I’m excited to meet Abby at some point. Final question: What do you think it means to flourish in the age of AI?

BROWN: I obviously started my career as an engineer, and since building the company, I haven’t been as involved in the technical work as much as I would like to.

There was once upon a time when I knew every line of code in Abby’s stack.

So I think to flourish in the age of AI, I started working on a few of my own technical projects again on weekends, just wanting to kind of get back into it. I wrote my first line of code in over 18 months just a couple of weeks ago.

I know everyone’s saying it, so I don’t want to sound like everyone else and every other voice out there, but how fast you can build things today is outstanding. Now it’s about being able to read the code, understand where it’s hallucinating, and understand how all the systems come together.

So you still need good technical literacy to be able to build very sophisticated and capable systems, but you don’t have to be writing every line of code. Writing every line of code now just feels really manual and primitive. I think how fast I was able to build different things I was working on would have taken me maybe two or three hours on a weekend, whereas two years ago it would have taken me maybe a whole week.

So I think that’s pretty awesome.

EL KALIOUBY: That is pretty awesome. Great.

Grace, that was awesome. Thank you so much for joining us on the show.

BROWN: Thanks, Rana. I had so much fun.

EL KALIOUBY: I love Grace’s energy. As an entrepreneur, you have to be bold, and it’s definitely bold to want to see more Abbys than dogs on the street.

Humanoids are going to be part of our future, working and living alongside us. But that means they need to know how to interact with us, not just perform tasks. Personally, I don’t want some sci-fi-looking humanoid in my house if it doesn’t have any social skills. And I’m concerned that a lot of the leading companies building humanoid robots are not prioritizing social and emotional intelligence. This is why I love Grace’s approach so much. She is building this social operating system that helps humanoid robots better fit into our lives.

That’s our episode for this week. Thank you so much for listening.