investor-interview:-bluelakeInvestor Interview: Bluelake

Jamie Rutledge, Partner, and Ethan Werner, Principal at Bluelake, tells CB Insights how they view the market, customer needs, and their company.

Are Sanctuary and Apptronik the only two, or do you have other investments in the space?

Jamie Rutledge: Those are our two primary ones that we focused on, and we did those a couple years back. But I’ve been looking at robotics since university, and I’ve been building robots since the ’90s. I was a systems engineer at IBM and Lenovo and ran systems research for them for a few years.

Then I worked at Dyson, so I had a lot of engagement with the James Dyson robotics research group at Imperial College. We’ve been working in the field for years and years, and it’s a domain I have a personal passion for.

There’s a whole lot out there. And it’s funny, I lived in Beijing for several years working at Lenovo research, so I understand the Chinese side of this pretty well too.

What first motivated you to invest in humanoid robots, and at this time, in those companies specifically? Why did you choose them over others you evaluated?

Ethan Werner: It really just comes down to having certain advantages that we thought mattered more. Jamie has a pretty solid background in understanding what’s important, especially moving forward, and I think those companies were both in a good position.

Jamie Rutledge: First of all, I’ve done a lot of patents in my past, something like 75-plus, and I’ve co-authored pieces over the years. I look at things in terms of the world changing fast, and those with a couple of stakes in the ground are going to win. I kind of look at it like the mobile phone industry. Do you remember when Motorola was the mobile phone company? BlackBerry had all these things.

They all had patents, and they started fighting with Qualcomm, and then the whole Google transaction happened. But who had the intellectual property (IP) was the real driver of all that. I think that’s going to start happening again. It’s not going to be an open-source environment for everything, at least I don’t believe it is.

Sanctuary has a pretty good IP portfolio. I just don’t know if it’s going to outlast the market creation, so that’s something that’s always open in my mind. But the things we’ve been looking at are safety, design, usability, and cost. Those are the functional things we cared about when doing diligence.

The teams we’ve invested in look at things through those lenses, and how that’s going to work, especially in the West, you’ve got liabilities you have to manage and understand. That’s something we care a lot about, because we don’t want someone to get hurt.

But the real reason is that the market is enormous. As soon as you can substitute a robot in for a human, the applications are massive. I think when we were doing diligence, the estimate was about $15B by 2030, and probably between $15B and $40B. If we can sub these things in and get them settled, the market is enormous. It’s totally beyond venture scale. Those are some of the key items I looked at.

Ethan Werner: I just think that looking at some of these technologies, especially the ones that have been very focused on from a hype perspective, like robots, quantum computing, and fusion energy, we’re now reaching a point where technology is unlocking the reality of these things. Twenty years ago, robots were starting to mature. Now they’re building cars and already replacing jobs, and the labor market is one of the largest opportunities out there. It’s a brilliant time to jump into opportunities like this.

How do you both see the commercialization timelines?I honestly go back and forth on this. We’re seeing some very good pilots right now, especially in automotive, which is one of the stronger areas. Agility has paid deployments already. Apptronik is obviously pretty strong here. Boston Dynamics is looking good with their Hyundai partnership, or being bought by Hyundai, and deployment there. I don’t know myself: are we looking at 3 years until these things are commercially viable to replace people, with enough uptime and success rates to take low-level tasks? Or are we looking at a long period of incremental improvement, like autonomous driving, where we were promised something around 2015 and it’s taken over a decade? Where do you think we’re at: 3 years, 10 years, or more?

Jamie Rutledge: Engineering mind on me here. Honestly, what you have is a systems engineering problem, and then you have a cost problem. There’s also a compute problem that needs to get solved in parallel so it can all work in a safe environment.

From a systems engineering standpoint, it’s an engineering problem: I’ve got to lift a 30- or 60-pound weight, and I’ve got to keep backlash down. All these system-level problems occur in a humanoid, so it’s good engineering that has to be applied. But there are costs associated with that, and that comes down to manufacturing. Every second on a manufacturing line is counted, and when you’re doing millions of units, those seconds really add up.

I think the companies focusing on manufacturability are the ones we’ve made good bets on. I can’t talk about the specific metrics because of NDAs and the like. But solving those problems and creating a supply chain that supports a lower-cost build is going to really turn the crank. Then you have to supply it with compute that can actually learn, do real problem solving, and do work. Those things are very important.

Design language is also important to me. I was at Dyson, and at Lenovo we cared a lot about design. You want a device or robot that you’re willing to be around. It has to be something you’re comfortable having your 3-year-old sit beside, something you don’t think is going to hurt her or fall on the dog, as a lot of people like to say, or walk through the line and punch a kid. Some of the things we’re seeing in these videos are stupid.

But those are the real fundamentals: the supply chain, cost, and manufacturability. I’ll be honest, I’m public about the fact that Jabil is their partner going forward. As an investor and supporter of Apptronik, those guys know how to build product. They’re working together to close those gaps and get the cost out, because building the robot is going to be fundamental.

How do you get those supply chains right? Neodymium is going to be a big part of that, because you’re going to build your own motors and overall actuators. When you put them together, how quickly does that happen? All these things matter in the equation to get a product out that’s safe, cost-effective, and designed in a way you want to be around.

Those are the fundamental things, and I feel very confident in our bets. We’ll see how it all plays out. But the cool thing is the market is $15B by 2030 at minimum, in our view. It’s huge. There’s a lot of room, and there are going to be a lot of players. We won’t be able to make enough robots. It’s going to be like the PC world of Gateway, HP, and IBM. There was room for everybody, all producing PCs at the time. There can be plenty of room for all kinds of robots.

So to summarize, we’re probably at a point where the proof from the pilot programs will arrive relatively soon. But then you have to scale up, build the supply chain for the entire industry. The robots will also still need to learn specific tasks on the job. They might know some tasks and be viable for those, but they’ll need to learn others.

Jamie Rutledge: Different environments are a factor. They have to learn environments they’re unfamiliar with. If you take some of these robots and put them out in the wild and try to see if they can get around autonomously, they freak out. It just doesn’t go well.

Also, part of the commercialization factor is that about half of the Fortune 500 manufacturers operate at Six Sigma standards. To my knowledge, maybe you know better, the best humanoid can hit about 98% time and success. That’s awesome for a pilot program, but it’s not Six Sigma for production on a manufacturing line. I personally feel there’s a huge gap between where they are now and where they need to go. Getting to 90 or 95% may be a pretty quick process from a training perspective, but getting that last little bit might be harder. Do you think that’s a factor, or are you less concerned about it?

Jamie Rutledge: If you look at any engineering problem, they’re all gen one, gen two, gen three. That’s what happens. Each company, and I’m not talking about any one in particular, has been turning the cranks. Agility started with their very first bipedal robot running around in universities, and they learned from that. The team was like, “Hey, look, I can step in a hole and not fall over.”

Then every university bought one to teach students how to engage with a robotic system. That was their first market. From that they learned, grew, and built a better robot. Now they have a robot that can run around carrying totes. That process is happening, but it’s not going to happen tomorrow.

I think the same will happen with Apptronix, definitely with Figure and everyone else out there. There’s this area of process where they’re getting manufacturability, and operational performance will come with it. You’ve got to do guard-band testing and shake-and-vibe testing, where you beat the crap out of these things so the connectors don’t come off. Wiring harnesses really do matter. Batteries have to latch on, stay on, and not cause a fire under different loads. Those are engineering problems. Good design is going to happen, but it’s going to take failure, and multiple generations of these products.

Even the motor controllers, right? If you short a full H-bridge, it’s going to fry the core, so you have to put the right protections around it. Iterative development is happening, so I think maturity is going to happen. 

Ethan Werner: I think it’s more mature than we give it credit for. These robots have been in development for the last 30 years. If a company came out and was able to manufacture a polished robot to demand within the next year or two, I could envision a capable enough autonomous system. When we try to achieve perfection, or 100%, humans aren’t 100%. Humans are terrible at jobs sometimes. They make a ton of mistakes. The real question is whether it’s good enough, and today we’re very close to good enough. 

Jamie Rutledge: I don’t think Six Sigma will be an issue for long. It’s all test and iterative development that gets us there. Try this, try that. Take that arm off, put this new arm on, because we made modifications. We changed the resistors and capacitors, and now we’re rounding out the transient curves so the device’s lifetime is higher. There are ways to go do that. It’s good engineering. It’s not revolutionary to get that sigma higher. As engineers, we know how to solve those problems. It’s a system-level problem, and we just go do it. I’m less worried about that.

How do you view the cost structure? A US employee has overhead, 401(k), and all that, probably $70,000 or more, while a humanoid might cost $100,000 to $150,000. But you get 2 shifts a day, 6 days a week, maybe, with uptime factored in, and you spread that over 3 years. In theory, a humanoid could be half as efficient, or a quarter as efficient, as a person and still make economic sense. Is that a correct way to look at it from a cost-structure perspective?

Jamie Rutledge: There are several ways to model it, but I can guarantee you they’re all wrong. There’s one thing I can tell you: we’ll need repair shops set up across the country to support these. At one of my former employers, we’d sell vacuum cleaners, and a customer would have a failure they couldn’t figure out. Maybe something’s clogged. They’d hold it up to the phone, turn it on, and say, “Here’s the diagnostics.”

We’re going to have a lot more sensors on the robots than on those devices. But then, what do I send in to get it fixed, because the customer can’t figure out how to solve the problem? So it has to go to a hub for repair. That infrastructure is going to have to be put in place.

Actually, I think Tesla is in an interesting spot, because they already have infrastructure across the country to do those things. They’ll just set up a new facility, maybe 5,000 square feet, associated with the robotics. They’re going to expand quickly and have service and support. The other robotics companies will have to build that capability, especially in these first couple of generations.

But yes, I think looking at a human and replacing the human, you’ll probably also get a better product on the back end, because the robot won’t fatigue. Think about a manufacturing line in the US where the robots are making the robots. Agility talks about that, and everyone is kind of talking about having their own robots as their first customer, on their own robotics line. They work through those bugs and solve those problems. Then you have a really interesting dynamic.

But what I was trying to say is that in manufacturing facilities, there are jobs where people get fatigued. They’re putting things in. I remember walking those lines as an engineer. We were making 8 million of these units, and you’d have 2 lines lined up. In Foxconn City you’d hear the presses going, the whole place shaking and moving, and you’d have a line of 20–30 people, each doing their little piece. Each one is a potential source of fatigue and a potential problem.

If you have a robotics line that can do that, and do it in a dynamic that can be reconfigured tomorrow, we’re running HP today and IBM tomorrow, it changes the whole dynamic. I think about the hands and manipulation, and as training gets better and better, those robots won’t get fatigued. They can manage it all. So where the parts are in the bin, they move them.

It really is going to change the world. Where we manufacture things doesn’t necessarily need to be in Shenzhen. It can be anywhere, and I think that’s going to be good for humanity. It’s one of the reasons, in the back of my mind, that I want to make these kinds of investments: to help propel us forward as humans.

Ethan Werner: I do think it’s pushing us forward technologically, but there are going to be some bumps in the road ahead. Bumps is definitely an understatement. If in the next decade these robots become capable and cost-effective enough to replace workers, even just in the US and in other countries with a lot more of that type of labor, we’re going to see some gigantic shifts in how the economy and the labor market work, and we need to deal with those. Robotics, humanoid robotics isn’t just a Roomba coming onto the market. It’s redefining the labor market as a whole. It has so many layers. 

I feel like it’s among the railroads, the internet, and refrigeration: core technologies that change how we work and how we live.

Jamie Rutledge: That’s the thing, though. It’s not going to be straight-up humanoids. Those probably won’t be the kinds doing bipedal manipulation and walking around. They’re all going to be on pedestals or the like, because they give a stronger foundation and less of that shaking.

There will be a multiplicity of different types of robots with different capabilities, and they’re going to be configurable and reconfigurable. That’s what a manufacturing line needs. It doesn’t need things that are rigid and stuck in one place, especially for smaller product portfolios that need hands and arms doing things. The compute has gotten to where we needed it to be to do those things. It’s really an amazing time in history.

There are areas where wheels or track systems could be cheaper, faster, and more mobile, since a lot of factories and warehouses are flat. And there are areas where a humanoid could be stationary and save money by not having a lower torso. Do you envision basically the same body, where you change the end effectors and the transportation source, so there’s a lot of flexibility?

Jamie Rutledge: 100%. Some of them may not even have a dynamic pedestal. Maybe there are 2 dynamic ones on the line, while the other ones are stationary, so they may be augmented, because it’s a cost-sensitive environment. These lines are very cost-sensitive.

I remember one line I was auditing, and I asked, “Where do you get these buckets?” They said, “Oh, we 3D printed them.” Well, not exactly. They’d done an injection mold and printed them, because shipping them here was too expensive, so they made their own buckets locally. Those lines are very cost-effectively run, and it’s going to matter.

Ethan Werner: On a longer time scale, which is how you should think about investing in early companies, humanoids make sense, because they’re designed for a world built for humans. When we built a factory or manufacturing line with humans in mind, we have certain constraints. The humanoid robotics market is filling the gap of a building built for humans.

But if we start thinking about things being built today with the future in mind, we can have purpose-built situations. That’s why I personally feel humanoid robotics will be just a small wedge of the overall robotics and automation market, because there are many situations where a purpose-built design makes more sense from a lot of angles. There’s still that last stage where you need a human involved to take those roles.

Jamie Rutledge: There’s something we haven’t talked about yet, and that’s safety and security. The automotive industry really put in the redundancy we needed to learn how to build things. Tesla has a lot of that infrastructure in place with their automated driving and the like. But you need different barriers to entry. The infotainment system of a vehicle has a very different security classification, or ring-fence, than the emissions systems or what actually makes the car drive or steer.

We’ll probably see a lot of that same kind of safety move into humanoid robotics, with redundancy built in. You might have a processor like an NVIDIA chip that handles everything under load, but maybe there’s a lower-weight, lighter-end processor that takes over if there’s a failure, to prevent a fall. There are all kinds of things we’ll start to see as these get more reliable, like in a hospital or a family environment. There’s still some innovation to be done here, and I’m really excited about it. So I just wanted to put that in there.

How do you view the humanoids coming out of China? Some are really affordable, less than $10K, and not really made for industrial purposes. But others are advertised for industrial use at significantly lower cost than some Western options. How does their cost compare with Western humanoids, and do you think we’ll have a bifurcated market, where Chinese humanoids sell into China and its allies, and Western ones sell into Western nations?

Jamie Rutledge: The crystal ball is unclear. But the recent Federal Communications Commission (FCC) regulation changes are interesting, and those will create a bifurcation. I think a lot of that is fear over data, safety, security, and privacy. That data is very transportable. Depending on how administrations form over the next 5–10 years, a lot of this will be driven less by product and more by governments, in my mind.

But the products we’re seeing today from a systems-design standpoint are very different. They’re usually lighter, smaller, with less carrying capacity than the industrial-strength ones carrying 40, 50, or 60 pounds. They’re different design points, built differently, so there’s a different cost. When you have an arm that weighs this much and you lift it, the backlash causes rocking in the whole system, so there’s a different cost matrix. It really depends on what you’re trying to do with the robot, how it’s applied, what markets you’re selling into, and what problems you’re solving. It’s all systems design, focused on what you’re trying to do with it.

As for bifurcation, I think a lot of it comes down to privacy and where the data is going. Industrial will be the same. If I’m GE making jet engines, do I want a humanoid pumping data all over the world, or one running on-premises with a clear delineation of where the data is, because I don’t want it to get out? We’re running into a very different dynamic.

Imagine a robot with 7 optical sensors looking at everything. Even in your home, you’ve got a Roomba or what have you on the floor, scanning around, seeing different things. It’s very different when a robot can go over and thumb through your files and look at your credit card number. It’s a very different dynamic.

If you aren’t already a client, sign up for a free trial to learn more about our platform.