Anyone Can Build a Robot That Works Once

The hard part is the 100,000th time, in the snow, when nobody’s standing there watching.

By Adam Mohamed, CTO & Co-Founder, Asylon Robotics

I’m a purist engineer at heart. I wanna make new stuff. I wanna be at the bleeding edge of what’s next. That’s what gets me out of bed every morning. I come in to work and ask, “What new thing are we gonna make?”

I came into MIT thinking I was gonna make spaceships. And then there’s this little drone thing on the side, and we get into it, all three of us, and we started messing around with them. They weren’t even called drones back then. They were called “remote pilot automated vehicles.” SpaceX was just taking off at the time, literally and figuratively, but there’s this other space that’s evolving. So, I’m a rocket scientist and said, let’s go apply that to drones and see what happens.

We turned the novelty into a workhorse

New technology usually shows up as a novelty. Something cool for a demo that ends up forgotten on a shelf. We took the novelty of a drone and turned it into a workhorse that operates all day, every day, rain or shine, hot or cold, Hawaii to Indiana in winter, and everything in between.

There’s a real need to take robots and put them in places where humans candidly aren’t that great. At 2:00 in the morning, sitting in the cold, trying to watch for someone coming by is not something we’re designed to do. We’re designed for dynamic events, rapid response. Not to walk the same path every hour, day in, day out, for weeks or months on end. Robots are.

The 100,000th time is the whole game

Anyone can make a nominal product that works in the lab. It’s 70 out and sunny, the robot works perfectly. Does it work perfectly when it’s snowing and 30, or foggy? I call it graceful degradation. As the thing is failing, it still operates to some degree instead of just falling over. That’s where the blood, sweat, and tears come in.

Getting Swap to consistently take a battery out of a drone and into a bay to charge, back and forth all day, every day, was a nightmare. Doing it once was hard. But doing it a thousand times? Man, you should’ve seen us jump for joy when we got it in for the first time. And then we said, okay, we have to do this hundreds more times. Let’s go do it.

We used it for 20,000 flights, and it had a 94% success rate. By almost every metric, that is phenomenal, except when it comes to flight. 94% means we’re crashing once a week with no human intervention. And if it’s crashing once a week and someone has to go recover it, you can’t pull the pilot from the site, which means it’s not truly an autonomous system. So we had to go build our own. We got it over 99.9%. And now we’re talking about something that could be feasibly deployed reliably in the field.

Build it, hand it to the operators, get the last 20%

I love that we’re vertically integrated. There is “no build a product and ship it out to some random customer.” It’s build the product, hand it to the team that’s gonna go run it, and the folks who maintain it sit right next to them. So we make the product, ship the product, run the product, and iterate on it faster than everyone else, because we know our product; we see it every day.

We take products to 80% and put them in operators’ hands, because that’s how we’re gonna get the remaining 20%. The easy 80 is real. The last 20 is that hard, and the only way to get it is data from the field.

Things break in ways you’d never guess. The unknown unknowns are just that, unknown unknowns. Sometimes the computer just says I’m cooked. I’m done. When those pop, it’s like Christmas for engineers, because it’s, crap, how the heck did this break? So now there’s a mitigation: if it loses comms with its own computer, pull the parachute, save the drone. It’s on all the drones now, and it hasn’t happened since. I’m risk-prone, but I’m safety-conscious. I lean forward on my skis, but I don’t go without skis in the first place.

This is the golden era of robots

I promise every engineer I hire the same thing, and no one’s called me out on it in the last 11 years: you will never be bored. Your work will always be interesting. I can’t tell you what it’ll be, but it will always be interesting. I liken it to the 1930s, the golden era of aircraft. This is the golden era of robots, where we get to do the coolest stuff, try amazing, interesting things. And things do not make it out of engineering. We’ll make bets on products, and guess what? They’re not good enough. We’ll look at it and say, cool. Great technology. Really cool try at this. Shelve it.

It is not a solo sport

I will take a team player every day of the week over a rock star that can’t play nice in the sandbox, every time, without fail. I will say no to the rock star, because if they can’t play nice in the sandbox, you’re not on my team and you’re not rowing in the same direction.

We almost do ourselves a disservice in engineering school; we don’t teach teams early enough. You can make an app by yourself. Congratulations. With AI, you can do that too. But we’re making hardware that does real things in the real world, and that is not a solo sport. It’s an integrated team. And that’s the real reason a robot we build works the hundred-thousandth time, not just the first.