Robotics’ Quiet Revolution
Or, Why the Robotics Industry is about to Explode
Over the last decade, billions of dollars have poured into robotics and autonomy. Despite all that money sloshing around though, you can still count on two hands the number of robotics teams that have successfully commercialized a product.
But quietly, outside the spotlight of the hype cycle, the robotics industry has built the scaffolding for a massive transformation. Why are things different now? Well, the answer is obvious if you look to history.
Robotics has been Incomparably Harder than Software
After I wound down my autonomous trucking startup, Starsky Robotics, I kept busy by consulting at software startups.
The first thing I noticed was that, at least compared to my own startup journey, their world just seemed so … easy.
At Starsky, just launching a “proof of concept” took 30+ engineers from 5–7 technical disciplines. Compare that to these software startups, who could support thousands of users with only a handful of engineers.
And they could move so fast! Serious moves like rolling out a feature took a few weeks at most. For Starsky, something as simple as switching truck manufacturers could upend almost everything we’d built, miring us in months of re-engineering.
Then suddenly it hit me.
It’s not that software is easy. It’s just that software startups only need to refine one small corner of the industry — because most of the stack has already been built.
Why Robotics is So Hard
Contrast that with robotics. For Starsky, completing a single test of unmanned highway driving meant building an entire robotics stack from scratch.
In addition to the autonomy itself, we had to build our own teleoperation, drive-by-wire system for the vehicle, controls tuning, hardware abstraction layer, and fleet management software. And don’t forget the APIs to interact with it all.
All of this was hyper-customized to one model of truck, doing one task, for a singular set of conditions. Changing any of those variables meant a top-to-bottom redesign of the whole darn thing.
Building robotics = building the whole stack
That sounds kind of crazy, right?
If you’re building SaaS these days, you wouldn’t even consider building something like hosting, payments, or authentication in house (unless one of those was your actual product). You’d have your pick of off-the-shelf tools. Each of those horizontal layers is mature enough to be considered its own industry!
And swapping out a piece of your stack? A little painful, sure, but probably not a full reengineering job. Because you likely haven’t built your product to be dependent on any given tool.
You would never build an entire software stack from scratch for a single application. Image source: TechTarget
With robotics, there’s no “other tool” you can swap to. Your only option was what you yourself have built. Which is tortuous and convoluted, like a giant Rube-Goldberg machine that completely fails if you remove a single part.
Robotics today is like Software was in the 1980s
Here’s the thing, though — the software industry took decades to get to the point where a hacker can stand up an app in a weekend.
Back in the 1980s, in fact, software development was pretty similar to what I’m describing in robotics today.
Most software was completely vertical — developed for the enterprise as giant, one-off builds. Often, companies even had their own operating systems, which their software had to be built specifically for. This made everything really complicated, really expensive, and finicky as all hell.
That all changed with MS-DOS.
MS-DOS was one of the first operating systems that became widely available across both consumer and enterprise machines.
The result?
Rather than having to rebuild applications for each company’s unique OS, developers could just build them for MS-DOS. Suddenly, they could focus on the application itself, without having to care about the underlying hardware or architecture.
Robotics Hasn’t Reached the Tipping Point
So when I say robotics is harder than software, what I mean is that robotics hasn’t reached its MS-DOS moment.
Oliver Cameron (who founded Voyage, now part of Cruise) said it to me like this: “there’s a tradition of building everything in robotics. It takes a lot of discipline not to just build your full stack in house.”
Like software in the 80’s, robotics today means building an entire stack yourself, making the applications complicated, expensive, and brittle.
The result? Companies like Waymo and Zoox have to raise billions in funding just to get to something that kind of works.
This is why robotics projects without insane amounts of money can seem like a pipe dream. And why, despite trillions of dollars sloshing around, the tangible progress in the industry feels lackluster compared to the poetic visions.
But There’s Hope
I hope you can see that if robotics today is similar to software in the 80’s … the amount of opportunity in front of us is almost unimaginable.
For software, MS-DOS was only the tip of the iceberg.
A common operating system paved the way for the unbundling of the entire stack, and unleashed a tidal wave of value at every layer. We went from a world of brittle, hyper-custom enterprise IT stacks to one in which a simple group chat app can be a billion dollar company.
And Microsoft itself? Turns out being the foundation of a massive technological transformation is a pretty good business.
The Great Unbundling of Robotics is Upon Us
If robotics can find its own MS-DOS moment, developers will be able to start unbundling the stack, kicking innovation into hyperdrive and unlocking tremendous business opportunities.
(I’d be remiss not to note the efforts behind ROS, the open source “robot operating system” that’s paved the way for much of modern robotics. But while ROS is the default for roboticists, it’s unapproachable to casual users — similar to UNIX or LINUX in the OS analogy. An important building block, but the tools that will truly accelerate robotics development are yet to come.)
Of course, some shrewd folks in the robotics space have already started biting off pieces of this pie. Like Foxglove, building a debugging tool for robotics applications. Or Formant, with their GUI for robot commands and monitoring. Or even Applied Intuition, the newly minted unicorn building simulation-as-a-service at shockingly attainable pricing.
But I believe that, like the software industry in 1992, robotics is still in the earliest days of a mass transformation.
As for me? You can probably guess where I’m placing my next bet. If you want to hear what’s next when I’m ready to share, follow me on Twitter @stefanesa.