Robots are becoming cheaper to make and more powerful because of AI. In the climate tech space, they’re already laying transmission lines, inspecting wind turbines, and installing solar panels.. And with labor productivity stagnating, immigration restrictions tightening, and the cost of labor rising, they’re looking even more appealing.
So where might robotics have the biggest impact on climate tech?
In this episode, Shayle talks to Andy Lubershane, a partner and head of research at Energy Impact Partners (where he’s a colleague of Shayle). Andy also recently wrote a blog post on the effects of autonomy across climate tech. They cover topics like:
- How more affordable parts and better foundation models are making robotics cheaper
- The high CapEx and low OpEx that make automation expensive to start, but valuable with high utilization
- Robotics-as-a-service companies that help to overcome these initial CapEx challenges
- The most promising applications, like manufacturing, construction, and maintenance
- The hopes for more humanoid general-purpose robots — and the challenges in making them
Recommended resources
- Steel for Fuel: Autonomy is real now
- SemiAnalysis: America Is Missing The New Labor Economy – Robotics Part 1
- Steel for Fuel: From SaaS to Robots
- F-Prime Capital: State of Robotics
Credits: Hosted by Shayle Kann. Produced and edited by Daniel Woldorff. Original music and engineering by Sean Marquand. Stephen Lacey is our executive editor.
Catalyst is brought to you by EnergyHub. EnergyHub helps utilities build next-generation virtual power plants that unlock reliable flexibility at every level of the grid. See how EnergyHub helps unlock the power of flexibility at scale, and deliver more value through cross-DER dispatch with their leading Edge DERMS platform, by visiting energyhub.com.
Catalyst is brought to you by Antenna Group, the public relations and strategic marketing agency of choice for climate and energy leaders. If you’re a startup, investor, or global corporation that’s looking to tell your climate story, demonstrate your impact, or accelerate your growth, Antenna Group’s team of industry insiders is ready to help. Learn more at antennagroup.com.
Transcript
Stephen Lacey: Hey, it’s executive editor Stephen Lacey. Many of you were loyal listeners to a show I created and co-hosted for eight years called The Energy Gang. Well, we’re back Jigar Shah, Katherine Hamilton and I are excited to announce our latest project, a weekly news round table called Open Circuit. Every week we’ll break down how major projects come together, how deals and policies get structured, and what it takes to build critical infrastructure at scale, all through the lens of current events. You can subscribe to open circuit wherever you listen to podcasts or follow it@latitudemedia.com.
Tag: Latitude Media: podcasts at the frontier of climate technology.
Shayle Kann: I’m Shayle Kann, and this is Catalyst.
Andy Lubershane: Everything it takes to build a robot is much cheaper now and you can build them much better with AI
Shayle Kann: Coming up: The robots are coming.
I’m Shayle Kann. I lead the Frontier Fund at Energy Impact Partners. Welcome. So lately I’ve been spending a lot of time learning about a broad set of technologies that I’ve been grouping under the term physical AI. These are applications of AI to the physical world. So AI for atoms, not for bits. To the extent that there’s going to be a big impact of AI on the energy sector, on climate tech, apart from its obvious impact on energy consumption, which we’ve already talked about AD AUM here, this area, physical AI is where I think the impact is going to come from. So we’re going to start talking about it more here. We’ve already covered it a bit. For example, when we talked to Doge from Google DeepMind a while back about AI from materials discovery, which is a category I think is interesting in the physical AI world. But one other obvious manifestation of physical AI where we’ve been exploring and investing at EIP for a while now is in robotics, which I find to be fascinating and incredibly interesting, as does my partner Andy Luhan, catalyst regular and EIP P’S head of research. Andy’s been on a bit of a robotics tear lately, so I figured I would bring him on to talk about robotics in general, but also its applications to the sectors where we focus. So here’s Andy.
Andy, welcome back.
Andy Lubershane: Thanks. I think this is my first catalyst pod of the Brave New World of 2025.
Shayle Kann: What a world it is. All right, we’re going to talk about your favorite topic of the past few months anyway, which is robotics. Robotics is not a new thing. It’s not a new thing in the sectors we’re going to be talking about. So I want to start with a why now. I think there’s kind of an obvious trite answer of like, oh, AI showed up, and so you could do more with robotics and capabilities for getting unlocked. I do think we should talk about that too, but just AI arrived. Why do you suddenly find robotics so interesting?
Andy Lubershane: AI is definitely one of the reasons that I’m interested in it, but that’s partially just because I’m a little bit of a contrarian. And so when people started getting excited about large language models and other generative AI stuff, a couple of years ago, I sort of started thinking, first of all, I’m not as excited. I’m personally only very recently have started to see applications where I think there’s some real, I won’t go ahead and say transformational value, but next level value in LLMs. So I sort of started casting about looking for where in the wide world of AI do I think there actually are transformational opportunities for energy and climate technology specifically, and robotics increasingly started to stand out To me. I think that’s both because empirically we’re seeing interesting companies and opportunities that are related to robotics in energy and climate tech, but also because of some big, I think big secular macro tailwinds that are really blowing in the favor of robotics.
And that’s both on the demand side and the supply side. So on the demand side, I think it is largely related to how difficult it is to get people to get labor, to do the kind of really hard physical work in manufacturing and construction to build all of this stuff we’re going to need to build and install in the energy transition. And that’s kind of related to three big macro factors that aren’t really going anywhere fast. One is the big global demographic transition. So especially in advanced economies, there’s just fewer young people who are being born today and were born twenty, twenty five, thirty years ago to do all that hard physical work. The second in some advanced economies and now in the United States is just the likelihood of a more restrictive immigration environment, which has been one of the primary ways that the US has escaped from this global demographic challenge that some other advanced economies haven’t in the past decade or two, is importing young people via immigration.
And then the third is this long-term trend that’s really concerning for economic growth overall, which is just labor productivity, kind of stagnating in advanced economies, not across the board. Obviously we’ve seen big improvements in labor productivity in certain sectors and basically in tech and anything connected to the internet and computers. People have gotten a lot more productive in the last 10, 20, 30 years. But again, in the sectors that we care about the most when it comes to decarbonization and really anything having to do with energy, which are basically manufacturing and construction, labor productivity has been flat really for the past 10 to 15 years. And in construction, it’s actually worryingly declining really for the past decade. And so we need people to be more productive in the physical world and robots are one way to make them more productive. So that’s the demand side. What do you think?
Shayle Kann: Yeah, I mean, one thing that I found interesting, clearly we’ve heard you and I have heard over and over again that labor is becoming or has been a constraint in a bunch of these sectors, construction in particular, but manufacturing and stuff as well. You hear it a lot anecdotally. One thing I haven’t heard though, I don’t know if I would realistically expect this, but one thing I haven’t heard is somebody saying, I can’t build X because I was not able to find enough people to do it, or I couldn’t cite this manufacturing facility. I failed the binary question of was I able to build it or not was no. So I hear a lot of the labor as a constraint, it’s getting harder and harder, I dunno about you, but I haven’t yet heard the labor was the cause of a failed attempt to build something.
Andy Lubershane: No, that would be really scary. I haven’t heard that either. I think the way we see it play out is rising wages, which is good for people who are doing the physical work, but more expensive for any kind of project that takes a lot of labor. And then again, like you said, just sort of generalized complaints that it’s hard to find people or that people don’t stick around for very long. They go and find work elsewhere, and so you have to train up new people. So yeah, I don’t think we ever see it in terms of project canceled because workers all quit or there aren’t any workers to do the project, but it’s sort of just a general drag on everything probably that is harder to pin to any kind of specific instance.
Shayle Kann: Okay, so that’s a demand side and I think, dunno, it doesn’t feel like maybe with the exception of restrictive immigration, which could actually provide a significant shock to the system on the labor force side, the rest of it is sort of ongoing, somewhat long-term trend. And so you could imagine there being more and more interest in automation across a variety of sectors, but nothing that would trigger a wave of deployments absent something changing on the supply side, the actual capabilities and costs of the automated systems in the first place. So let’s switch from the demand side to the supply side. What gives you a fair amount of optimism on the supply side, meaning what we can produce for robotics and the cost of it.
Andy Lubershane: The supply side is definitely where the more interesting why now elements come into the picture. So we can start with ai. AI is getting better, it’s getting better a lot faster. And that’s not just related to generative AI building large language models and large multimodal models and doing digital stuff emerging from that ecosystem. We’re also starting to see some real seminal research on the application of generative AI to robotics. People should go and check out things like Google’s Palm E model that was published I think a year and a half or so ago. There’s this research group at Stanford that built a robot they call mobile aloha using new methods of training. This very simple, relatively low cost basically system of a couple of robotics robotic arms on a little push cart. They’ve made really remarkable progress at the ability to train robots to engage in less structured environments and perform really highly unstructured tasks, which has been the challenge in the past for anything in robots.
I mean there are hundreds of thousands, millions now of industrial robots installed worldwide, which are largely just these robotic arm systems that are doing largely deterministic tasks in factories. So passing one part from one side of an assembly line to another and doing a very deterministic weld for example, between two sheets of metal that’s been going on and actually gradually growing amidst these macro demand side tailwinds really for a couple of decades now. But what AI makes possible is to much more quickly train robots to navigate in more complex chaotic environments where you can’t train them to do the same thing over and over and over. And that’s where a lot more of the most exciting opportunities in energy and climate tech come in, both in manufacturing, but I think more in the deployment side where you’re actually having to build stuff out in the real world, which is inherently chaotic and unstructured and impossible to program deterministically.
And so that’s one thing. AI is making real progress and I do think unlocks robotic opportunities that were not available even five to seven years ago. And then in addition to that, one of the things that I think has excited me is just it’s getting a lot cheaper to build robotic systems of basically all kinds because all of the components of building a robot, all of the inputs into just about any kind of robot are getting cheaper. So again, starting with ai, the computing hardware to run AI is getting cheaper. All the computer chips, GPUs have gotten much more highly performant in just every dimension very rapidly and continue to, but also the more mundane or relatively mundane aspects of a robot. So all of the kind of sensors you need to navigate in a less structured environment. So lidar for example, the cost of lidar systems has come way down, I think 90% in the past 15 years. And that’s largely been driven by investments in autonomous vehicles, which are now finally coming to fruition and becoming a real commercial enterprise. But along the way, they had to invest a lot in making sensor technology much cheaper. Other boring things like actuators, just electric actuators for the joints that make robots move. I’m sure no one listening to catalyst needs to be told how much battery costs have come down in the past 15 years. So basically everything it takes to build a robot is much cheaper now and you can build them much better with ai.
Shayle Kann: Well, that’s the thing about robotics in general. When you start looking at these companies who are doing various things in robotics, but this is almost universally true, if you’re able to price the service at roughly the equivalent cost that the human labor alternative would’ve cost because the cost of the robots is so low. Now the payback periods on these things are remarkably fast, assuming you get high enough utilization. So the big questions are can you price at the equivalent or close to the equivalent, how much is the thing being used? But with such low CapEx, you get these extraordinarily fast payback periods. And so the unit economics for this stuff just looks remarkably good. And so the question is, will that pricing power hold one and two, will they work? Will they be reliable, at least as reliable as humans, if not more reliable? Can you trust them? Will your customers be happy with you? But because that bomb cost, the cost of the equipment, the materials that goes into it is so low, you don’t need crazy high pricing. There’s no premium on a robotic based system to replace a human labor based system. Generally speaking,
Andy Lubershane: I think that’s right. And actually there’s a subtle connection here to batteries and battery costs and battery performance and electrification in general. And you mentioned utilization, right? So if you are building a robot, it is still substituting CapEx for opex effectively, right? You’re taking opex, which is human labor costs out of the equation for some number of years, and you’re investing a big chunk of CapEx upfront in order to do so in terms of building the robot. And so because of that in general, utilization is really important for the techno economics of any robotic system. You want to get high utilization to get that rapid payback period or IRR and battery technology coming down has a couple of impacts on that economic equation. One is that yes, the CapEx in general just becomes cheaper because the power system for the robot has become much cheaper battery and a motor.
And secondarily, all of the other operational costs generally tend to fall the cost of energy. If you’re running a robot off of an onboard engine, it’s going to be a lot cheaper if you’re running it with a battery. And that favors higher utilization, generally speaking, electronic components, motors for example, relative to an engine, electric actuators relative to hydraulic actuators, they’re also more, they also have lower operations and maintenance costs. They’re more durable. And so running these systems electrically tends to be able to get you the kind of high utilization and low maintenance costs and low operational costs that you need in order to make that equation pencil. So in a way, actually electrification, which I know is a theme that you guys talk about a lot on catalyst, but electrification is really good for robots as well. They sort of go arm in arm.
Shayle Kann: Yes. Although I would say the two things that these are not enough to stop the tide stem, the tide of the robotic revolution, but the two things that sort of point in the opposite direction to me that are related to what you just described are one, electrification is great and robotics is a form of electrification, but when electricity prices are rising, it’s harder and electricity prices are definitely rising. So that’s one. And two is this CapEx opex thing, which I hadn’t really thought about, but it is a good point. It is similar to not just batteries like renewables. Renewables are relative to fossil generators are all CapEx and very little opex as compared to fossil generators, which are some CapEx and some opex. But the challenge of the high CapEx low opex switch in today’s environment is that interest rates are high and when interest rates are high, the cost of financing stuff is higher and you have to finance more if you’re trying to get a payback on your CapEx.
So both of those things are probably marginal headwinds I guess against the robotic revolution. I haven’t heard anybody saying, well, I would automate this system, but the cost of electricity is too high, just again, because unit economics otherwise are so good. But you could imagine that being a factor. And similarly, we have seen folks who are having to finance deployments of a fleet of robots of one kind or another, and the cost of capital does matter to them. So either of those could be transient and you could see even more tailwinds in the future if they go away. But those are the two things that I think from a macro level make it, I dunno, marginally more challenging. I think
Andy Lubershane: I agree, especially on cost of capital. I think that’s the bigger constraint and that is why we see there’s this idea of a RAs model, a robotics as a service business model, which shifts the cost of capital from the customer to the robotic producer. And actually I think that’s generally the way we’re seeing some of these systems get rolled out. It also helps because sometimes the robot, a robotics company can make sure that their product is utilized more if they’re able to deploy it amongst a bunch of customers rather than just sell it to one customer. So that takes the edge off. But agreed, cost of capital is a concern, one of the reasons that it’s so important that the CapEx just the cost has come down so far.
Shayle Kann: Okay, so we’re talking about robotics in a very generalized sense, and you’ve mentioned building stuff, but I do want to talk about it specifically in the context of energy and climate, what we spend our days on. So talk about where you think there’s the greatest applicability, the most promise for this new wave of robotics in let’s say energy in particular.
Andy Lubershane: Yeah, so as I said at the beginning, I am increasingly cautiously optimistic about generative AI in the digital world. Large language models I think are starting to show some real promise making gains across lots of areas of knowledge work, which does apply in just about any enterprise in the energy sector for example. But it is super hard to get excited for the implications for energy and climate tech in terms of the big picture. And so generally speaking, I think where robotics come into play and where they can make a real difference is in manufacturing and construction and getting down to brass tacks. What we’re starting to see, and I know a bunch of us at EIP are starting to see in the market are these very use case specific. I think the investment firm F Prime, which puts out an annual robotics report, which is really good, calls them vertical robots, right?
They’re robots that are built for a very specific vertical sector, really in kind of an increasing array of applications that are relevant for climate tech. So there’s a number in our portfolio at EIP that we can point to. It’s a company called InVision, for example, that has a robotic system which consists of a quadcopter drone, kind of like you see all over the place these days, but it’s a somewhat specialized version of that, but even more importantly, connected to this ground-based system, which enables the drone to string a transmission conductor and be a tool to help facilitate, make much safer and faster the process of stringing transmission lines, which as we know is one of the bottlenecks in terms of cost and speed in the energy system today. And we think that they can make a real difference there. So that’s one example. And really these robotics companies range far and wide because increasingly, again, they’re able to operate in much less structured environments than the industrial robots we’ve seen in the past.
Shayle Kann: One thing I want to point out that’s interesting about InVision is people think of robotics as being a one for one replacement for a human doing a task, and InVision is a replacement for humans, but humans in helicopters, basically helicopters doing tasks. And so one of the things that I think is interesting there is if you’re looking for applicable use cases, it’s like where, yes, there’s a bunch of expensive manual labor but human labor, but it’s also where we’ve just dramatically over-engineered because that’s been the only way to build something in the past. And that’s where you can get these much more modular bespoke, much more intelligent, faster solutions to produce the same outcome.
Andy Lubershane: It’s simultaneously over-engineered like a super expensive person in a helicopter and under engineered in the sense that you really don’t want to have a helicopter flying anywhere near a transmission tower. I mean, there are real safety implications there, and I think there have been people that have been killed over the years installing transmission conductor in those settings. And actually I believe that’s part of the reason that InVision was founded actually in Australia, is to avoid those kind of safety incident in the future. So yeah, that is another potential benefit of some robotic systems is you can take a human out of a dangerous situation.
Shayle Kann: Alright, so InVision’s a good example. There’s a couple of others that we’ve invested in but haven’t announced yet, so we can’t talk about in great detail, but maybe paint a broad brush as to additional categories, whether we’ve invested or not that you think there’s sufficient opportunity in climate energy.
Andy Lubershane: So in the past 10 years there’s been a lot of experimentation using drones. Again, the sort of standard quadcoptor form factor to do kinds of inspection work across lots of different types of big and tall assets. And that includes, again, lots of stuff in the energy sector, wind turbines, for example, inspecting a wind turbine with a drone is a fantastic use case because yet again, as in the case of InVision, the alternative is sending a person to climb a hundred meter tower or higher and attach themselves to a giant wind, turbine the cell and climb out on the blades connected by ropes. This is dangerous, difficult work. It’s hard to find people who have the stomach for sending a little drone up to collect imagery of that turbine is a fantastic alternative and that application has gotten a lot of traction, but again, as robotics technology improves, especially the AI side, which makes these systems capable of interacting with their environment in more complicated ways, we’re seeing a transition just from doing inspection work to doing actual maintenance work.
So now there are a handful of robotics companies that are doing maintenance on wind turbine blades, for example. Another area that we’re seeing robotics come into play is in the construction of renewable assets. So there’s also a handful of companies out there that are showing a lot of promise in large scale solar project installation. That’s another area where you have to find people, lots of people who want to go out into, in some cases the middle of nowhere in a typically hot sunny place and do extremely repetitive work lifting heavy objects. I think average kind of solar panel now weighs something like 80 to a hundred pounds.
Shayle Kann: Is that right? Yeah, I think it’s the new ones because getting bigger and bigger, making bigger and bigger modules. So your 700 watt module now is 80 plus pounds, which as you can imagine, it now takes two, often three people to lift it. And you’re taking it a long ways actually out in the field. You have to see it to really understand it, but there’s just a lot and that’s just the pick and place of the module. But all these systems also have tracker systems and there’s actually a lot of work to be done to install the tracker, install the panels on the tracker, aff, fix them, all that kind of stuff. So it is, it’s a remarkable feat of engineering and construction that we build these multi hundred megawatt solar projects, but it is an incredible amount of labor.
Andy Lubershane: It really makes my shoulder hurt just thinking about it. And a solar project you can kind of envision as the perfect frontier, I think for robotics today, because the work is very repetitive. I mean, you’re doing largely the same thing over and over. In some cases over thousands or tens of thousands of acres with thousands of these solar panels, the exact same systems, but the terrain is complex enough, there’s enough differences in how you have to approach installing the racking system in slightly hilly terrain or depending on the moisture content of the ground and installing the panels on a tracking system themselves. You can’t completely automate that in the same way you do. So in a factory, there’s enough chaotic elements that means you really do need the next level of AI and robotics in order to automate things effectively. But it’s not so challenging. It’s not so chaotic that you need robots that are capable of responding to anything that comes at them,
Shayle Kann: Right? We know one of these companies did a pilot out in the field in somewhere in the southwest of the United States, and it was a one month pilot, and over the course of that month, they experienced every type of ambient condition. There was everything from 110 degree day to high winds to sleet to ice, just the moving out of the factory and into the world. If you’re trying to operate as close to 24 7 or at least every day as you possibly can, they’re just ambient conditions you have to deal with in addition to undulating terrain and whatever else it might be. So yeah, I do think of there being a spectrum here of on one hand you have a robot sitting sedentary in a factory doing the same thing over and over and over again with very few things that can screw it up. That’s probably the easiest application of a robot. On the other end of the spectrum, I dunno, is maybe a fully autonomous vehicle that has to plan for every possible edge case in the real world and has enormous safety implications every time it tries to do anything. And that’s a really, really hard problem, which obviously we’re making a lot of progress on thanks to Waymo and folks, but this kind of field construction type of thing is somewhere in the middle of that spectrum, I think.
Andy Lubershane: Totally. Yeah. I think the question you have to ask if you’re considering a robotic solution is just like how long is the tail of weird stuff that might come up that it might have to deal with? And in the autonomous vehicle case, that tail is extremely long. You’re going to encounter some weird shit in a city in San Francisco driving along the streets, everything is sort of an exception, whereas in a solar field there’s a lot of exceptions to the norm you’re going to encounter, but it’s manageable to be able to build a robotic system relatively quickly. That can take a lot of significant human labor out of the equation. And I think the more that roboticists are looking for these applications across the energy world, the more they’re finding them and developing robotics solutions. So we see them in vegetation management. For example, we have in our portfolio company at EIP company called SI Motors, which makes an electric autonomous commercial lawn mower for mowing big commercial lawns.
We’re also seeing some specialized versions of that that are being developed specifically for large solar projects where they have to sneak under the panels and do vegetation management, which actually is a significant expense, one of the larger operations and maintenance expenses on large solar projects. We’re now seeing robots applied to an unfortunate area of climate tech, which is wildfire mitigation. Everything from robotic systems that are clear cutting and burning areas of overgrowth near, for example, electric power assets to prevent ignition to companies. I saw recently a company that’s using a drone to go in and doing thinning in forests both for forest management and collection of woody biomass for various purposes. So it’s hard to say anything. It’s hard to make blanket statements about where the opportunities are because I think we’re just going to see them popping up more and more in these vertical robotics applications.
Shayle Kann: Well, we’ve been talking about so far mostly are host of companies. There are tons of ’em now that are applying robotics to a specific vertical and doing generally some version of that. Robots as a service type model. They’re various versions of it, but generally that is what they’re doing. They’re not selling systems, usually they’re operating those systems and selling a service, but that’s not really the only category of novel robotics companies. And so I’m interested how you’ve been thinking about where the opportunities lie in the application of robotics into energy and climate. Is it in these find use case build robotic system deploy system for that use case types of businesses? Or are there others that you think are interesting as well?
Andy Lubershane: I do really like the vertical robots, the find use case, build a robot to solve that use case or a collection of related use cases world. And I think what we need to see from companies that are taking that strategy that they can get off, I think for companies that are adopting that strategy, they need to be betting on a significant first to market or one of the first to market advantage. And what I mean by that is they need to be able to get their units, their robots out in the field early and start to learn about those kind of strange exceptions that they’re going to be encountering in the field, start to collect more and more data so that they can train their robotics model to do things more and more efficiently. So that two to three years in when another company comes along and thinks to themselves, Hey, robotic components have gotten a lot cheaper, and I think I could build some sort of robotic system to solve the same problem or similar set of problems, that company A who entered first will have enough of an advantage and come down the performance and cost curve fast enough that they’ll be able to maintain that lead and that there’ll be sort of the obvious choice for potential customers.
And I think we are at this kind of inflection point in the robotics market where it is now possible to build a solution fast enough to get it into the market fast enough and actually get it into the field fast enough to generate that kind of a data moat and a training moat and just sort of a standard sort of learning curve. So that’s why I think that space is pretty rich and really seeing companies developed there for new use cases in particular the other places you could make a bet. I think at the other end of the spectrum, and there are plenty of, there’s a lot of bets or a lot of capital being made are a lot of capital being put into bets. In this side of the spectrum is the general purpose robot and the marquee version of that is the humanoid robot.
And the theory of the case there is that the world that we live in is the world that humans built and therefore lots of tasks are well suited for a humanoid form factor. And that goes for both in controlled settings like factories as well as in the field. And there have been at least in the hundreds of millions of dollars invested now, I think in some cases in the billions of dollars invested in individual companies that are developing humanoid robots with that theory of the case that they will be more general purpose and can apply to a wider range of tasks. There’s a bunch of reasons why that is. I think at least one order magnitude, maybe a couple orders of magnitude harder problem than building these kind of specialized robotics as a service. Obviously it’s a problem that some companies think that they have a good chance at solving Tesla being sort of the most preeminent example.
Shayle Kann: Right. Okay. So those are good categories. I think there’s maybe a couple more worth highlighting for a moment that are not specific to energy or climate or anything, but just out there in robotics world. One is the generalized models, the equivalent of the LLMs, but for robotics, some of those are popping up right now and also raising a lot of capital. And then the other category maybe at the other end of the spectrum is just components for robotic systems selling bits and pieces and bobs that are going to go into robotic systems under the assumption that we have a big growth market. And so if you’re the best at selling whatever widget that needs to go into every robotic arm or something like that, then you’ve got a big business. How do you think about those two categories?
Andy Lubershane: These are the picks and shovels for robotics, the inputs to robotics. On the AI front, there’s this category of robotic so-called robotic foundation models, which are kind of trying to be like an LLM for movement, a large movement model. There’s a variety of ways to try to describe it. Then there’s also a category that’s related, which is sort of other software tools for accelerating the development of robots. For example, simulation software which tries to build in realistic physics engines so that you don’t have to collect data from a robotic system moving around out there and interacting with things in the real world that you can actually collect lots of that data virtually in a simulation before you actually deploy things physically, which could make a real big difference because otherwise, unlike in the case of training large language models where you have the entire corpus of the internet to work with just this mind boggling amount of data out there.
If you’re trying to collect data to train a robot, kind of the only way to do it is to go out and have the robot go and do stuff. And that just takes real clock time and clock money that the LLMs have not had to deal with. And so I think the next generation of simulation software could be really valuable in this case. I don’t know about the robotic foundation model world as much. I have my skepticism about that approach kind of in a similar way as I have some skepticism about LLMs themselves, which is that I’m not sure about the long-term moat, and I’m also not convinced that the companies who are building robots, especially really, really highly sophisticated robots are willing to seed that part of the value chain to a third party. So curious about it, but come to it with a bit of an eye of skepticism. And then like you said, there’s all the non-software components too. There’s sensors, it’s perhaps maybe we’ll get to that next generation of lidar, cheaper and better lidar sensing solutions. Maybe it’s some other sensing modality and there’s actuators and motors. Robots are in many ways sort of more complicated autonomous vehicles. They have to have a lot of the same sensing systems embedded within them. They have to have actually probably more of the actuation systems because if you’re building a humanoid, it has a lot more joints than a standard car.
So yeah, I mean that may be an area where that the best bet is to make on the existing suppliers, and it very much is sort of a boring but not so boring picks and shovels bet on the robotic revolution.
Shayle Kann: Alright, well I think we’ve covered the first level of robotic obsession. There are many levels beneath that into which we’re both diving at the moment, but can cover another time. So Andy, thanks again for your time. As always,
Andy Lubershane: Always happy to be here and always happy to talk about robots. It’s one of the things that it’s most palpably exciting. Everyone can think about a robot and see how it might solve some real problems and do things that you just don’t want to do.
Shayle Kann: I’m picturing middle what I imagine middle school Andy to be like. And I imagine he was probably into robots.
Andy Lubershane: Yeah, he’s not so far away.
Shayle Kann: Andy Lubershane is my partner and our head of research at Energy Impact Partners. This show is a production of Latitude Media. You can head over to latitude media.com for links to today’s topics. Latitude is supported by Prelude Ventures, prelude backs, visionaries, accelerating climate innovation that will reshape the global economy for the betterment of people and planet. Learn more@preludeventures.com. This episode was produced by Daniel Woldorff. Mixing and theme song by Sean Marquand. Stephen Lacey is our executive editor. I’m Shayle Kann and this is Catalyst.
