Inside the Startups Bringing AI to the Physical World

Welcome back to TechMates. Today we are cutting through the sci-fi hype to see what robots can actually do in the real world. We've got three founders tackling totally different environments. We'll cover why a robot folding your laundry is still a massive headache, why autonomous surgery is still actually easier than self-driving cars, and how robotic submarines could save our cities from running out of water. First up is David Ings breaking down why your trusted dishwasher won't be replaced by humanoid robot anytime soon. Let's go and do it and see what happens. Ideas are worth like jack [ __ ] Absolute jack [ __ ] Like if you can't execute, there's nothing.

Just take that initiative to actually take whatever crazy ideas that they might have and just try to pursue what comes of it.

Don't be afraid of ambition. Chase it. Love it. Wrap your arms around it and and say it out loud.

Kiwis don't know any better. They will just get up and give it a shot. That's why we were first to get to the top of Mount Everest. Go to dream that ultimate dream. Like what is it that you ultimately want to do? And it's all achievable. Do you already see that? Or when do you think that will happen? Where like you have a robot that sort of just give it a vague task, right? In the same way that you give an LLM like a vague task. And then sort of just like goes and reasons and then outputs like does some series of actions.

And because we have seen so much hype around what robots could potentially do over the last 10 years and what they really can't do. Like even at the moment, the reliability on a robot waking up, getting the photo, and making its way home alive is still really low. And it's like, "Come on, you know, you're seeing all these videos of these amazing robots doing these amazing things." And it's like, "Hey, just to build base reliability into basic logic is still very, very hard." And then people say, "Well, hey, when is this robot going to, you know, reason?" So I I say like, "So So just like ground robots have like exponentially more complicated than aerial robots. Aerial robots generally always have GPS coordinates.

It's very easy to get a robot to go from point A to point B in the sky is you know, exponentially more complicated indoors when you don't have GPS for a robot to know where it is and to get from point A to point B. Plus things can get in this way. There's people moving around. It's like significantly more complicated. And then you start adding in something like, you know, when is a robot going to be able to fold the laundry? So so even without the reasoning part of hey, go go figure go find the laundry in the house and just like fold this. You know, it may land up on the couch folding a blanket or something stupid. Like even knowing the context of laundry laundry basket. It's clean laundry, not dirty laundry. It's just like there's massive problem. It's It's like really really complicated.

So So I think that is the holy grail and I think that's where the embodied AI is really focusing on to say like yeah, we can actually make a robot that because it looks like a human and people look at it moving and they freak out and they go this thing is now a human. It's like no, it's like 0.1% of a human right now. You know, what's more of a human is chat GPT. That's way more of a human. This this robot is like not a human. It's super dumb. So So yes, as I say, even finding the clean laundry is going to be a mammoth task, never mind picking it apart and putting it down and folding it and then knowing where each piece goes intuitively and opening drawers without them falling out or you know, what if there's not enough space in the drawer.

That is going to take a long time to solve and so the question with any of these robots is you know, a lot of these humanoid robotics companies learned from Boston Dynamics who started off really generic saying we're a cool tech company, we can build robots and then we'll figure out what they're good at. A lot of the humanoid robotics companies are saying look, let's absolutely focus on a use case that's that's simpler. It's really simple. Health and safety environment like a factory context for a building a BMW. You can put a perimeter tape. We're not going to hurt anyone. Check. We don't have kids getting hurt by robots folding laundry. And then it's like highly contained use case. Like we can pre-train it on every object it's ever going to come into interaction with. There's only three types of panels, car panels you're ever going to touch. Here they are pre-programmed ready to go.

You're only ever going to stand like in two places there and there. We can map the environment. So So they've constrained the domains of problem right down and get those working and then they can build layer layer on layer on layer. So even when these robots start coming into the the home, they's going to start off with incredibly basic tasks. And I think we had a chat in the car around even cooking, you know, like even the best cooking robots, you know, at best can make one type of pizza or at best can make soup. That's like the extent of their of their capability. So So I think is AI going to provide these incredible leapfrogs where year on year we're going to make massive progress. Man, I hope so, but history history says no. History with, you know, autonomous driving says things take way longer than we think to get a sock folding robot that can put the socks away in the drawer.

That's And what what do you think will be the first kind of mass, you know, application of kind of robotics in in a consumer setting like in in a home for example or in a workplace?

The big question is still the specialized robot. It's like, "Oh, well, I could use that robot to vacuum my floors." Like, "Well, just get a vacuum cleaner robot. It's way cheaper and much more reliable." Well, maybe it could wash my dishes. Like, you already have a dishwasher, you know, but it needs to be packed. So I think you very eventually realize, well, outdoors you're probably going to get a lawn mowing robot. You're not going to use the humanoid there. Indoors, you already have a washing machine and a dishwasher and certain things. And then if somebody said laundry folding, maybe it's just a laundry folding cabinet which they have now which, you know, they're working on. And all the thing that thing does is fold laundry, you know, out comes from the dryer, goes in that thing.

So So I think it is a fascinating thing and I think the problem is the robot's going to have to be incredibly generic to actually be useful for those kinds of environments and you might find specific robot form factors just outperform it for a while until it becomes really, really good and it can pack the dishwasher, maybe maybe cook, you know. And I in. Like when it can start doing that stuff, all of us are going to buy one, but it's It won't be like like a Judy from the Jetsons, but it'll be more like, you know, I talk to like my chat GPT or my Siri on my phone, and there's like 10 specialized robots in my house each doing a certain thing, kind of thing. I think it's going to be all of those. I think we will continue to see specialized robots getting cheaper and better at mowing your lawn rather than teaching a humanoid robot to drag your lawn mower around.

But then also, I think there's just a level of task which requires high mobility and height, you know, which is designed for a human. So, to get dirty plates off the table, take them to the kitchen, rinse them off, stick them in the dishwasher, pretty hard to build a robot that can do that that doesn't look like a human. That's designed for a human humanoid robot. But I think we are going to see a number of those things shrink. So, you know, by the time they're ready, like some of their tasks already would be done by a specialist robot that's going to continue to be just better than them. Mhm. Maybe laundry folding, you know. Right. My my sense is like I feel like and I don't know if this directly answers that question that I asked, but it feels like military use case might be the first first time where we see like at real scale like um autonomous robots, right?

Like you're seeing it in Ukraine now with with the drones and a lot of the drones are becoming more and more autonomous. And we looked at a couple of companies that are that are doing that. Basically just yeah, killer drones, right? Like And again, it's highly contained. I think one of the problems is people always associate robots with batteries not not not petrol. And if you look at Robotics Plus, it's like hey, if you're trying to last in a farm for a long period of time, maybe using diesel is the way to go. Maybe it's a genset that powers a battery, but it's, you know, hey. So, so I think the question is what is, you know, is it is it then an autonomous vehicle not a robot?

I think the biggest problem a lot of people, you know, like just that use case of using a spot robot to chase sheep around, it's like, you know, if you look at these farm dogs, they they'll operate for 10 hours non-stop. This robot will last half an hour out in the field, you know. Um and it's going to step in the first puddle and then its battery's fried. And it can't jump it can't jump fences. So, you have to keep opening fences for the stupid robots. So, so imagine trying to take a robot like that into warfare. It's like it can hardly carry anything. It's only going to last a couple of hours. You know, what are you going to carry a whole bunch of gensets around to like charge your robot fleet? It's like, you know, it's okay with a drone it can last 3 hours it can get to its target job done, you know? But but some of these other robots is like there's real power problems.

Uh there's real practical application problems. And yeah, so so you'll find hey maybe maybe a dog robot who work with a great company in oil and gas industry called Ghost Robotics. They they build a version for the military which does perimeter security. So, if you're on a base, robot wakes up, walks the perimeter, make sure nobody's trying to climb in, goes back and charges up and keeps going. Yeah, great use case. David makes a great point. Robots work best today in highly constrained spaces like a taped off factory floor. That perfectly sets up our next guest Nick. He's the co-founder of Andromeda building autonomy for surgery. Why he's doing that because performing surgery is actually an easier problem for robot than navigating a city street. Let's see how Nick is bringing cognitive assist into the operating room. What was the grand vision and where did you get started?

So, it's interesting right my co-founder and I had the same concept and a roughly the same concept independently before we met. And so we we had both thought about autonomous surgical robots from different life experiences we've been led to this point. And so for him he built the first autonomous truck that ever drove on a freeway without a person. And then he also like like myself had some tough experiences where we both had companies that got to a really good like high valuation great great point. In my case it was two different companies that both both did that and then you know, it wasn't necessarily a smooth path to the goal line from there. And then for him he also had a situation where this company was super hyped up did great things but then didn't work out in the end. So, yeah he he had thought about okay what else can I do? I've I've tried the driving autonomous driving thing."

He's a robotics background, a lot of autonomy work, then was thinking, "Okay, what else Where else can we replace human labor with robotics and autonomy that's going to be valuable?" And he had thought that surgery is a really fruitful place to do that, that this is a place that autonomy can add a lot of value. And then I had thought about just all the problems I had seen. I'd probably been in hundreds of operating room cases, different companies, and being in labs and whatever over over my years and just said a lot of things go wrong in surgery. And as a meta need, we should do something that caused there to be fewer complications. Seeing I had also kind of seen Waymo's launch right before this and Tesla was building self-driving and just thought that this is this is actually easier problem. Surgery is a lot easier than driving. There's It's a more constrained environment.

Yes, there's anatomical variability. There's more than most people think actually in the anatomy variation from one one person to the next. But still, there's the world has a lot fewer features than the driving world, which is basically the like you can counter anything that exists in the whole in the whole world really when you're driving. Then there are other agents that are acting against you when you're driving, too. There's a lot of, you know, other drivers, there's all kinds of pedestrians and bikers and animals and whatever else might get in your way and construction and there's endless endless number of things. But surgery was easier. And so, I had thought about, "Okay, I can This technology exists. We can build this and it would solve these problems that I'm always seeing in the operating room and make surgery much better and have a huge impact on human life."

So, we met up and had both been thinking about this and then decided to to brainstorm on it. We met up through the YC network. And when we looked into this, we were we basically realized that this is we were we were convinced this is this is definitely going to happen. Like if we don't build this company, we will see somebody start this company in the next 10 years if not sooner and it's going to be huge. It's bigger like what Intuitive is now, which is worth like almost 200 billion now, this could be a lot bigger than that. Because this is going to add a lot more value intuitive robot is great. I mean, it adds a lot of physical value to surgeons, but adding the cognitive assist is much more impactful. You can potentially allow surgeons to do a lot more procedures in a day. Uh so, that's adding value, you know, to hospitals. And then also have much better outcomes.

And so, that's, you know, adding value to the the patients. And then it also There's a lot of value to the surgeon as well. And so, what What is the website? What does intuitive do? Just for context. They have So, their main product is the da Vinci robot, which is a I think it's forearm abdominal surgery robot that does a lot of the like abdominal procedures. Like prostatectomies is one that urologists do. And then you can do gallbladders and other other things that are happening in that part of the the body. Right. And how does it how does it sort of help? Like why do surgeons use it? So, this is uh Got a picture of it here. So, what is it? Yeah. How does it help? Yeah, I mean, it's So, it it makes the procedure physically easier. It allows them to be They basically control It doesn't really change the controls that much. You You use the controls in this little console.

And then you can do the procedures in the in the body with these arms through the robot. And it does And it One One reason it does help is you don't have to cut Like it replaces a lot of open procedures, at least initially did. So, where people would cut into the abdomen, now they can use these different robotic arms to do it instead. So, there has been some There's always a lot of criticism that da Vinci doesn't improve outcomes. Now it's been shown it does improve outcomes in certain procedures. Though there's a lot more It's only scratching the surface of what you could do with an autonomous robot.

And this is what the This is what the surgeon actually does. Surgeon's actually just sitting there.

They're in the console. Yeah, they're in this console and they're operating through that. And these little uh actuators they have mimic the tools they would already use for what's called laparoscopic surgery, which is There's a non-robotic version where they have these tubes going through the abdomen that's not open. So, they're not cutting through. You're using these little tubes. And in those laparoscopic procedures, the surgeons manipulate these these little controllers that are kind of like what the da Vinci robot has. So, there is some precedent for this. Not being robotic though, with the robotic system they get really good, you know, dexterity and then also good visualization. I see.

So, you're saying laparoscopy, you're actually manipulating the tool yourself with your hand versus this that you're there's there's like you're kind of disconnected from the actual like things you're manipulating. Yes, exactly. Interesting. Okay. And then And you're saying

a mainstay in in most major hospitals in the US and a lot of the world. I mean, this this is everywhere now. And you're saying this is great, but like let's add some intelligence to it or something. Is that right? Yeah, so in what we what we realized that the physical challenges of surgery are not limiting the outcomes that much. It's why It's why DaVinci though it has shown some outcome improvement has not shown a lot. That there's a lot more you could get from using AI to assist surgeons. And I think everybody even intuitive understands that. Most people think that it's 20 years away that we're going to be able to do this, but you know, as we've talked about, we can build it now. And so, that's what we're doing. We just said we've got a great opportunity to build this now. The technology is definitely there. MedTech is always 10 years behind.

And so, we have 5 10 year head start on when this probably would be built otherwise. And so, we should just do it and let's do it as fast as we can and just try to expand autonomy and expand the number of procedures we offer as as fast as possible. Kind of crazy to think that we might see autonomous robots in surgery before we see fully self-driving cars everywhere in the streets. But while Nick is focusing on the operating room, our final guest is tackling a massive crisis deep underground. Globally, 30% of our fresh drinking water is lost because of pipe leaks. Let's see how Harrison and his team deploy autonomous robots into high-pressure water pipes to find those leaks before cities run completely out of water. So, you see sort of like you you servicing contracts for like utilities.

And then separately, you having like a platform for other for other companies to sort of build on to on the hardware platform for their niche applications. Is that kind of the right way to think about it? So so I mean so you're both kind of like a service company and a platform like why not just do the platform part? That seems easier. I mean like have it other people use your platform to to do the service piece themselves.

Yeah, and I think that's that's definitely where we've we've started and and what's kind of the next step for the robot, but I think it's much harder to achieve the same level of impact with with just the the platform play. To get fleets of robots that talk to each other that are constantly living in pipes. Yes, that's that's not easy. That's that's definitely the moonshot goal, but it also has the most potential in terms of impact. Globally 30% of all our fresh drinking water is lost to pipe leaks. That's you know, 4%

How much? 30% That's the global stat. In in places like Europe where they've got really old pipes, it's it's higher. Like Mexico City where you know, the city is sinking so they have so many problems with their underground infrastructure. The the shipping bringing tankers full of water in because they can't maintain it. So there there definitely places that it's it's much worse than that 30% figure. Here in Sydney, I think it's about 12 13% which is you know, better than that global average, but you've also got places that are you know, constrained in terms of their water supply. I mean a couple of years ago South Africa almost ran out of not South Africa, sorry. Cape Town in South Africa almost ran out of water. Um and that's just going to How do you How do you run out of water? Yeah, they they came really close to what they called day zero.

Their reservoir their dam was being used at a faster rate than it was being replenished and what that means is as you know, climate change continues to worsen in certain regions and water insecurity grows, there will be more and more cities that go through dry periods and use their entire water supply before it gets replenished. Um that was kind of almost the case when when Purelink was conceived. We were under water restrictions where our Wivenhoe Dam was getting low. And so it's something I think might be uniquely Australian in the sense of a we've grown up around it. We've we've had water restrictions. It's the land of droughts and flooding rains. And so we see that juxtaposition of we can run out run out of water. It's a thing that can happen. Doesn't happen in New Zealand, I guess.

But the I guess the bigger vision of like having these uh kind of autonomous robots or fleet of robots swimming through the pipes, what are the biggest kind of engineering challenges to to get to that place? Yeah. So there's there's three key challenges that you need to solve. The first one is a form factor that can navigate and withstand that environment. So you're looking at about 120 PSI is the standard, you know, water pressure that that you would experience in a water main. Sounds like a lot, but it's about the same as about an 80-m water depth. And there's hobbyist, you know, submarines that go down to that level. So that's that's okay to deal with. The next part about it is autonomy. It's got to be able to make decisions inside of the pipe with no human operator human in the loop control. So that requires, you know, significant amount. Why is that why what Why is that important?

Well, if you want economies of scale of fleets of robots out in these pipes finding and fixing leaks before they become a problem, you would either need to scale a team of humans to control each and every one of these pipes, which is quite difficult to do, or you facilitate that through autonomy. And I think that's where robotics is heading in terms of the efficiencies that it can introduce. That's that's something that I think we have a unique advantage in as we start to roll out our robots into more pipes. We collect more data and our robots get better and better over time making decisions inside of these environments. And then the the third aspect of that is your power and communication.

So, your your power system, as I mentioned earlier, there's so much kinetic energy running through a pipe that's that's a relatively, I say simple, but it's a it's a relatively obvious one to tap into and then power you uh sorry, communication. You really have to use low bandwidth flags with really long wavelengths that can penetrate through the ground. So, you have to, you know, be able to indicate really simple things like leak here, blockage here, because you're not going to be able to stream out HD footage or provide a point cloud of of what's going on. So, it is really reliant on that autonomy to be able to make those decisions. Thank you for tuning in to TechMates. If you enjoyed this episode, be sure to subscribe, leave a review, and share it with friends. We'll be back soon with more stories of Kiwi and Aussie founders reshaping the future and disrupting down under.

Until next time, keep dreaming big and daring to disrupt.