Matic Robotics co-founder on the hard problem of housekeeping

In the late 2010s, Mehul Nariyawala observed founders across Silicon Valley racing to get into the autonomous vehicle space. Rather than thinking bigger, he went smaller.

“I’m like, wait a minute: There are 100+ teams who believe they can build a robot that doesn’t bump on a bump in a road, but they can’t build a robot that doesn’t bump in a home? That doesn’t make sense,” he told us.

So Nariyawala and Matic Robotics CEO and co-founder Navneet Dalal set out to bring AV tech to your kitchen floor. Turns out, that may have been the more difficult path.

“A self-driving car has its own challenges, insanely hard problems; I don’t mean to trivialize it,” he told us. “But they don’t need to go really, really close to another car. Usually you have about three, four feet of distance…versus in here, you want to go right next to the furniture, clean, but not bump it, and that’s a much higher-precision challenge.”

Nariyawala talked with us about his background, the inspiration behind Matic, the challenge of founding a hardware-intensive business, and what’s next in robotic housekeeping.

(We also took a test drive of a Matic. Check out our review over on Tech Brew.)

This conversation has been edited for length and clarity.

Tell me a little bit about your background. Where did you work before Matic? What were some experiences that got you thinking about founding your own company?

Navneet and I…have worked together for about 20 years now, and his background is in computer vision…In 2005, both of us met at this startup called Like.com, and that was probably the first computer vision startup in Silicon Valley. If you kind of take that time back, I think computer vision was still esoteric; feels a little bit like quantum computing today, where people may have heard of it, but no one knew what that was. And on my part, I didn’t really know anything about computer vision at all.

I really got very lucky in joining this startup on computer vision in a very serendipitous way. And I spent most of my time doing [the] product and marketing side of things. When that startup was eventually acquired by Google in 2010…we thought we had a good ability to go start a company together. That’s when we did our first company called Flutter, which was gesture detection.

While we had built really, really good technology, there was no business model; it was a really good app, but people weren’t going to pay for it. But luckily for us, both Apple and Google were interested in acquiring us, and we ended up getting acquired by Google. And then inside Google, just after we got acquired, Nest was also acquired, and from there, we got a chance to transition to the Nest team. 

What ultimately led you to found the company? Was there a moment that made you say, “I must start a robotic floor cleaner company?”

There were multiple reasons. One was that if you have Level 5 robots for cars, which means cars drive like humans, then why can’t we have Level 5 robots for home and the indoor world? And if you have them, what do they even mean? And we came to this conclusion that if Level 5 robots means they drive like humans, then Level 5 robots for the indoor world must be that they map like humans, navigate like humans, avoid obstacles like humans, clean like humans, manipulate things like humans, do chores like humans—then shouldn’t they have a perception system like humans?

Our entire indoor world is built by humans for humans to fit our perception system. So if we’re going to have a robot to do things on our behalf, computer vision and vision-based perception system is the right way to go. And that was our background. That was our core strength.

Second thing is, if you’re building self-driving cars, you have this amazing advantage in Google Maps, which tells cars where the road is going, and GPS, which tells cars where [they’re] located. Without these infrastructure pieces, cars would be lost no matter how smart they are. But then how would a robot know whether it’s on the right side of the couch or the left side of the couch? Whether it’s in front of a refrigerator or dishwasher? And that contextual awareness was missing, and again, we felt like vision was the right way to go, and we can solve this problem by literally giving robots eyes and brains.

Purely from a user perspective, as a father and homeowner, I want to live in a perpetually clean home with perpetually clean floors, right? [And] I don’t want to clean it myself. I don’t want anyone in my family to clean…[My wife] would always complain that [she feels like she’s] always cleaning, or you get mad at kids…and that’s because kids are sort of walking entropy generators.

Humans are a creative species. We’re not a repetitive species, and we spend way too much time inside homes. So then we’re like, how much time do we spend? Turns out, on average, a family of four will spend about 15 hours a week driving, but about 45 to 60 hours a week doing home chores. That blew our mind.

No one is really solving that problem. So if we can build a startup and build a product that not only is insanely helpful, but it gives people their time and energy back, that just sounded like something that was worth building over the long term, and we felt like this was the company, and this was the project on which we can devote the rest of our careers and not get bored, because it’s not just about floor cleaning. Floor cleaning is the first product.

How do you start from floor cleaning and maybe go to the next set of problems along the way? [How do we get to] Rosie the Robot, which we’ve been working on since the ’60s, but we haven’t even made any dent into it? So that was the thing: How do we get there? How do we build it? And we felt like floor cleaning was the right place to start.

What’s the most difficult part about building a business, particularly one that’s so hardware intensive?

If I knew how hard it was going to be, I would have probably never done it…If hardware is 10x harder than software, then robotics is probably 100x harder. Because not only do we have to build an entire AI layer, similar to, let’s say, OpenAI or Anthropic, but we also have to build hardware and combine everything as a systems problem.

If a robot were to bump into an obstacle, did it fail because the algorithms are bad? Did it fail because the camera decided to turn itself off, or be out of focus? Did it fail because the CPU is slow? Did it fail because the motor…failed to obey? What are the challenges? And that entire challenge is actually really, really hard to know…So you not only have to invent the robot and the algorithms, but we’ve been inventing tools and QA systems along with it as well, and that just took a really, really long time.

Was there a moment in the development of the Matic when you thought you would fail? And if so, how did you overcome that problem?

Throughout the entire time of building Matic, we never thought that we would fail to build a robot; that wasn’t a question. When was the question.

The reason being is that we knew the technology is there. We had seen the signs of our progress…But when it comes to robots, we almost always want to delegate, and the bar for delegation is much higher. So while the mistakes inside a home are a little bit more trivial than, let’s say, mistakes self-driving cars would make, realize that tolerance for those mistakes is much lower, because these are considered trivial tasks, and we just want to delegate, right?

The second thing we learned is that the indoor world is the most chaotic, unstructured and dynamic space, especially homes. So homes are actually far, far harder than people realize. And Google Maps for home doesn’t exist; GPS for home doesn’t exist. So everything a robot has to do in a very precise manner, and you can’t clean right next to the furniture without going very close to it. So in order to go close to it and not bump, you need much higher precision.

What’s next for Matic?

The first priority is getting the existing robot…feature complete. In 2019, we wrote down certain specs of this ideal robot vision that we had, [including] things like voice and gestures. So can we give you a very simple way to just give it instructions? Because many times, as a parent, we don’t have our phone with us, and sometimes things spill, so it’s like, “Hey, Matic, go to the kitchen. There is a stain near the refrigerator. Go clean it.” Can we give that instruction to the robot?

Maybe eventually there will be two, three robots doing different tasks inside, almost learning ones. And when that happens, and you can trust three robots, then there may be an opportunity to say, OK, here’s a three-in-one or four-in-one robot that does everything. So that’s how we think about it, that the next five years are very much around this purpose-built robot that does one thing insanely well.

I’m really, really excited about where the robotics industry is going…There’s a lot of investment, a lot of hype.

Everything that the world is talking about in terms of robots is absolutely coming; if is not a question to me, when is the question. So how do we productize it? And I hope people spend more time thinking about it as a product, because consumers don’t want robots. They want solutions to their problems. So what is the problem we are solving, why are we solving it, and how are we solving it? That is the critical piece of the puzzle.