Industrial vehicles should be automated.

Instead, they’re dangerous, often far away from population centers, and surprisingly hard to staff – which means the vehicles grow in size, become expensive to operate, and are less fuel efficient.

The problem with automating industrial vehicles is that robotics looks more like software in the 80s than like mobile app development today.

There's limited tooling, and everyone is building the full stack from scratch for each new project.

In contrast, the industrial companies interested in buying robots are highly optimized – and they need robots that can keep up by fitting into existing systems and executing specific tasks.

At Polymath Robotics, which comes out of stealth today, we’re building general autonomy software for industrial vehicles, so that technical teams can automate the physical world.

Our generalized basic autonomy system is designed to be optimized by technical teams so that industry doesn’t skip a beat.

We have vehicles driving unmanned almost every day and large deployments in process.

But technical teams can start building on top of us right now, for free, in simulation via our REST API.

Oh, and we’re in Y Combinator this summer.

Background

The last time I wrote a widely distributed blog post was at the end of something.

I'd just shut down my last startup, Starsky Robotics, and I spent a long time thinking about how I would do it differently.

Back then, a peculiar phenomenon would happen.

Nearly every month, a billion-dollar company would reach out to try to convince me to pivot.

"We’ll give you $7m," they’d offer, "and in return, your engineering team will spend the next 6-18 months automating a niche vehicle, for a task that no one else in the world needs."

That would have been impossible, because Starsky's stack was designed specifically for on-road trucks.

Any shift in that stack (vehicle, requirement, or task) would've required a major re-architecture and rewrite.

Autonomy is a deceptively challenging problem.

While there are many billion-dollar industrial autonomy applications, they’ve been impossible to service without re-building the entire autonomy stack each time.

Until now.

At Polymath Robotics, we’re building all of the basic generalizable autonomy needed to automate the 50m odd industrial vehicles doing work today.

We’re focusing on components necessary for autonomy, but insufficient for specific customer value – path planning, hazard detection, behavior trees, human detection, controls tuning, and safety.

Polymath abstracts away the complexity of building a demo-able autonomous vehicle, and leaves the niche-specific behaviors to you.

That means technical teams and select industrial partners can quickly deploy autonomy on-site at scale, no safety driver needed.

Progress

We’ve made fast progress since my co-founder, Ilia Baranov, and I started working on this last September.

By the end of 2021, we were already running unmanned tests with no person on site.

Today, we drive unmanned autonomous vehicles for 4-6 hours every day with nobody on site (often without any human actively monitoring them), and we have customers developing on top of our platform.

With Polymath, autonomy no longer needs to take 2 years of engineering just for a demo.

So today, I’m also incredibly excited to release Caladan : our autonomy agent commandable via REST API, and observable in simulation on an internet browser. No installation of Gazebo, ROS, or Linux required.

Example application

Let’s say, for example, that you wanted to automate a tractor to till a farm (which just means turning the soil before you seed).

Before Polymath, you’d need to get a tractor, retrofit it, build out autonomy, build a safety layer, get to the point where you can take the person out, and then build the front-end app that a farmer can use.

With Polymath, you can just focus on the app, and use our Caladan tool to connect to our API and command a virtual tractor in sim.

If you want, we can even let you command our physical demo tractor from your app via our API.

And when you're ready, Polymath software will seamlessly transition from the Caladan simulation to a real vehicle.

Want to see Caladan in action? We built out this exact use case as example code you can play with.

This progress wouldn’t be possible without my fantastic co-founder, Ilia Baranov (who previously built home robots at Amazon’s Lab 126, and ran research products at warehouse-robot pioneer Clearpath Robotics).

We’re joined by some of the smartest people I’ve ever worked with, from juggernauts like Hyperloop and GoogleX.

I’m also thrilled to be in Y Combinator again, this time with Polymath.

Y Combinator pushes companies to optimize for a metric.

When Starsky went through YC in 2016, we had no improvable “metric”, just the binary outcome of whether our tech worked by Demo Day.

At Polymath, we already have something that works (!!), and can now focus on serving users.

So – working on an autonomy project?

Sign up today to start building on top of our API.

Or, if your company operates industrial equipment, reach out and see if we can work together directly.

Let’s automate the world together.

-Stefan