The Washington PostDemocracy Dies in Darkness

Shared scooters aren’t always where you need them. A U-Md. researcher has a plan for that.

Researchers are trying to get the scooters to reposition themselves in more convenient locations

Derek Paley, director of the Maryland Robotics Center, recruited students to build a self-driving scooter at the University of Maryland. (Derek Paley)
6 min

In the late 1990s, Derek Paley was working as a software analyst for a Navy contractor in Northern Virginia when he encountered three words he didn’t know could be strung together: autonomous underwater vehicles. “This was the coolest thing I’d ever heard in my life,” he said.

Within hours, Paley approached the start-up business that was building them. He joined the company, went to graduate school to learn more about how to control the submarines and now is a leading roboticist at the University of Maryland, where he uses insight from animal behavior to better coordinate schools of the underwater craft.

But he also loves scooters and often is seen zipping around campus on a one-wheeled electric version. When he used to roll beside his bike-riding daughter to elementary school, “adults would either look at you like you were to be despised, or they just would ignore you,” he said. Kids would stare, speechless.

In addition to directing the Maryland Robotics Center and working on artificial intelligence for the Army, Paley has recruited students to build self-driving scooters. The group is trying to get the scooters to reposition themselves in more convenient locations so people will use them more often. Their work comes as many people across the Washington region are weighing how to commute for in-person work for the first time since early 2020.

More Q&As on modes of transportation in the Washington region

Paley spoke to The Washington Post about the project and its inspirations. The interview was edited for length and clarity.

How did you decide to make scooters drive themselves?

We did 101 customer interviews. Two students and I were on a lot of Zoom meetings and we got really immersed in the micromobility community. That includes everybody from shared operators like Skip and Jump and Lime and Bird to e-scooter manufacturers. We didn’t have a product or solution or anything. We’re just trying to understand what the problems are.

I’m an avid scooter rider, but in an unconventional way. I’ve been riding for four or five years now, like unicycle-style scooters that are electric-powered. People point at me on campus. I remember one time somebody said, “I think I’ve just seen the future.” That was pretty hilarious.

I have a pretty close working relationship with the [U-Md.] Department of Transportation Services. They have a vision for how College Park can change. The Purple Line is coming in, more and more buildings are going up. People are parking farther and farther away from the campus. And so the campus partnered with Veo, which is a shared scooter operator, to provide scooters. It’s not hard to recruit students to work on this. We’ve built two prototypes that are going to enable a scooter to operate without a rider.

What are some of the things you actually need to solve?

The basic discipline that I’m trained in and that I teach is dynamics and controls. Controls, or what we call automatic control theory, is at the heart of a lot of modern technology. It’s a concept that utilizes feedback to regulate a system to some desired behavior. A classic kind of boring example is a thermostat. It gets too hot, it turns on the air conditioner until it gets cold enough and then turns it off. That’s an automatic control system.

Bike-share and e-scooter companies, hit hard by the pandemic, may come back stronger

What we’ve done is we’ve taken two off-the-shelf scooters and equipped them with a lot of really expensive sensors. Our approach is, let’s start with a proof of concept and then we can make the electronics cheaper and we can dial back exactly what we need to add on there.

The students right now are working on a simple challenge I gave them, which is to drive from Looney’s Pub, which is on Baltimore Avenue, to the engineering building. I think it’s about 500 meters, give or take, if you follow the appropriate traffic rules. And there’s a bunch of different terrain there. There is a sidewalk, there’s a trail, there’s a bridge over a little creek. There’s a plaza in front of a building. That’s going to be our goal for this fall, to do that autonomous mission. Right now the scooter is a little wobbly. It can drive itself over short distances. But we’re still working on getting the GPS integration and migrating that outdoors.

Can you do this safely?

Our underlying premise is that you’ve got 2,500-pound self-driving cars driving at 55 mph in a lot of cities right now, and cities are bending over backward to bring that technology in. There’s a disparity in the safety considerations of different industries. I guess it just goes to show how bad human drivers are that we’re willing to entertain self-driving cars. But a scooter, even a heavy scooter, is about 50 pounds, and it only goes 10 or maybe 12 mph, so you’re just talking about a lot less kinetic energy than a motor vehicle. And if it’s driving autonomously, it’s not going to be going 10 mph. We’re not going to hurt anybody by driving it. It would be more of like an oddity that people would probably knock over just for fun.

What will the scooters do on their own?

We’re not trying to drive the scooters 15 mph across a city. That’s too hard a problem. I can’t solve that problem. There are insurance issues, there are regulatory issues, there’s just the pure complexity of it. We’re talking about microscale repositioning. We’re talking about 10 to 100 meters. There’s actually value in that. If you get one more ride per day because you move the scooter 50 meters or 100 meters, maybe out of the alleyway, maybe from the departure side of the Metro station to the arrival side of the Metro station, that’s a huge win for a shared operator.

Most scooter rides, at least for students, they’re like five-, 10-minutes long. And you’re only getting three or four rides a day. So the scooters are only being used for like 20 minutes out of 24 hours. That’s crazy to me how underutilized the scooters are, and so we think, if you had an app where you could summon a scooter just like an Uber and it would come to your building from the parking hub that’s 100 meters away, that’s going to make a difference in terms of your time savings. If we can help with adoption, I think we have a viable value proposition.

D.C. Council approves new e-scooter regulations

Another thing is this idea of augmented riding. Can you detect when somebody is riding on a sidewalk and automatically limit their speed to an acceptable speed? Can you add perception to a rider, so that it will beep or warn you, just like modern cars, if there’s a potential collision impending? All the same instrumentation we need to add to a scooter to make it autonomous can benefit a rider, too.

How does your research on “bioinspiration” inform your work on robots?

What’s beautiful about studying animal groups and other, what we generally call multi-agent systems, is that even if the individual agents are very simple and obey very simple rules, the behavior that emerges from the group is quite complex. Fish tend to reorient their swimming direction using the swimming directions of their neighbors. Birds tend to separate themselves from other birds using the relative position and velocity of their neighbors. Look at the beautiful patterns that you see sometimes in D.C. in the dusk when all the starlings are swirling around to roost in the Smithsonians. As an engineer, we look at those for inspiration and say: Can we replicate some of these capabilities in robotic platforms that are governed by simple rules?

How do you describe artificial intelligence to regular people?

The three main elements of a robot are basically sensing, thinking and acting. AI is robotics without the act part. AI is just taking data and processing it and thinking about it and making decisions, but a robot actually exerts a force in the physical world. It does work. Machine-learning is kind of at the heart of current AI. And machine-learning is the idea that if you feed an AI enough data, then it can actually start to detect patterns that humans maybe couldn’t detect, or it would be prohibitively difficult for a human to process all of the data.