Refraction AI’s Matthew Johnson-Roberson on finding the middle path to robotic delivery

Refraction AI calls itself the Goldilocks of robotic delivery. The Ann Arbor-based company, which recently raised a $4.2 million seed round and expanded operations to Austin, was founded by two University of Michigan professors who think delivery via full-size autonomous vehicles (AV) is not nearly as close as many promise and sidewalk delivery comes with too many hassles and not enough payoff. Their “just right” solution? Find a middle path, or rather, a bike path.

The company’s REV-1 robot, which co-founder and CTO Matthew Johnson-Roberson debuted on the TechCrunch Sessions: Mobility stage in 2019, was built on a foundation of a bicycle. At about 4 feet tall and 32 inches wide, the three-wheeled vehicle can travel at up to 15 miles per hour, which means it can stop quickly to avoid obstacles while still being faster than a human.

The intermediate speed also means that the REV-1 doesn’t need to see as far ahead as a full-size AV, which allows it to function well on radars, sensors and cameras instead of requiring expensive lidar, according to the company.

Johnson-Roberson has spent nearly 20 years in academic robotics. Universities are home to many of the advances in field robotics, but the average person doesn’t see many such applications everyday when they look out their window. This desire to make something that is useful to the general public has been a huge motivator for the academic-turned-founder.

The following interview, part of an ongoing series with founders who are building transportation companies, has been edited for length and clarity. 

TechCrunch: You unveiled Refraction AI on the TechCrunch stage two years ago. How has it evolved since?

Matthew Johnson-Roberson: It’s been a really exciting ride. At that time, we had one vehicle — the one that we rolled out on stage — and now we have 25 vehicles in Ann Arbor and Austin, which we just announced. So things have changed quite a bit in the intervening years. We had already predicted a lot of changes around food delivery, specifically, and lots of those were accelerated by the pandemic.

From a technical perspective, it’s super interesting trying to figure out how to produce multiple versions of these vehicles and get better at building things. And from an operational standpoint, we went through a couple of different iterations trying to figure out the right way of charging people money to deliver to them.

More Transportation Founders

We started out being direct to consumers so they would order from our app and get deliveries that way, and evolved that to working directly with restaurants, and just being a logistics layer, as opposed to trying to be another DoorDash. So yeah, all those things have happened in the intervening two years.

What about you? Are you still the same guy that you were when you first launched?

I think I’ve learned a lot, and yeah it’s been an interesting journey for me. It’s my first company. I’ve worked in autonomous vehicles since the early 2000s, so from a technical perspective, this wasn’t new, but man, it was just a total fire hose and unique experience trying to figure out how to do all this.

Me and Ram [Vasudevan, co-founder] — it’s the first company for both of us, and we’re professors at University of Michigan. So just getting the opportunity to figure out how to build a company was a really cool process. As we started to mature, we were able to bring in some experienced people to the executive team. We got a new CEO, who just started, and it’s really neat to work with people that have done this before.

Probably the most stressful part of the company was trying to keep everything going and raise money, and do all of this during the worst bit of the pandemic. That was pretty grim, but we had a lot of time on Zoom to do that stuff, I guess. But yeah, I’m super proud of where we ended up on the other side.

You mentioned the change in leadership. You went from CEO to CTO. Why was that? 

I really wanted to try to expand and scale the company, and I think one of the things that was really key there was understanding that we were severely understaffed for doing all of that.

Ram and I had never really scaled anything on the company side. I’ve been working in academia for quite a long time, so for me that was a great opportunity to keep working on the things that I find interesting and work less on the things that I don’t find interesting.

So that was one big piece of it. The second is just acknowledging how challenging a technical problem this still is. Trying to make sure we have enough resources to solve it and not spend all my time in meetings with investors. That doesn’t get your robot to move from point A to point B much better.

Let’s talk about the tech. How would you say it’s developed over these past two years? What’s different from your competitors?

It still boggles my mind that nobody has tried to copy what we’re doing. There were 10 or 12 sidewalk robot companies in early 2015, 2016 and 2017. Many of them, with a few exceptions, went out of business.

Three years ago, we were saying, look, full-size AVs are going to take longer than everyone is saying, and there are limitations around both sidewalk robots and full-size AVs that make them impractical for large scale urban delivery. None of those things have changed. In fact they seem to be truer now than they were back then.

We kind of looked at that and said, “Okay, I think we have to really come at this from a different angle.” And so the genesis of the company was understanding that’s not the answer and also that full-size vehicles are probably farther away than everyone’s saying. Those two competing forces were what prompted Ram and I to think okay it’s worth trying to make a company around this.

From a technical perspective, the big differences are that if you’re moving faster than a couple miles an hour on a sidewalk, you’re trying to solve problems that are much closer to full-size vehicles in a lot of ways, but then you don’t have the benefits of traveling in a full-size traffic lane.

Only 30% or 40% of the full-size vehicle stack stuff works, but the rest is stuff that we’ve had to design new and from scratch just because the problem of where you travel on the road creates a lot of different technical challenges. And while we’re still the only people that are trying to do that, it means that we’re trying to forge our own path.

You call yourselves the Goldilocks of AV delivery robots. How does this approach affect your unit economics?

We’re getting to this idea of what the challenges are with the sidewalks, and what the challenges are with full-size vehicles. I think the primary economic challenge with full-size vehicles is just the expense of the vehicle itself. The asset costs, when you add the sensors and the vehicle itself, are really high.

And the dirty trick of the gig economy is to take that cost, and you basically take the depreciation of those vehicles and stick it onto your gig economy worker, right? They use their own car, they bought their own car, you tell them that we’re gonna pay you X and then you’re in charge of maintaining and depreciating that vehicle yourself.

So that is a really difficult thing to think about: How you would take that model and do that with a full-size vehicle, this really expensive asset that you would have to own and maintain. The economic pressure from the sidewalks is that if you go slowly, it means you can do very few deliveries per hour, and to make up for that, you would have to make them very inexpensive relative to human beings.

I think we’re still at the point where those vehicles are not free, and running them is not free. It’s really difficult to figure out the economics of that and make that work for us.

Again, in this Goldilocks zone, it goes fast enough that you’re doing enough deliveries per hour. Then the cost of that vehicle is not astronomical because it’s going slower; it’s not a car, and you don’t need to see as far as long as you use cheaper sensors and all those kinds of things. All of that comes together, hopefully, to make unit economics that are favorable when you compare them to gig economy workers, which is really the central competition for food delivery in urban areas.

What is your strategy for convincing lawmakers to allow your tech on the streets and also to encourage it?

Surprisingly, two years ago we thought that was gonna be a much bigger hurdle, and it has not been a huge impediment. All the markets we’ve been in have been very friendly. Both the city and the state have really been not only open to us, but encouraging us to run and expand both in Michigan and now in another state that we’re announcing soon.

The problem is, it’s very difficult to find a jurisdiction in the United States where you cannot run a vehicle without a driver in full autonomy mode these days. Like almost everywhere you could; the problem is, it doesn’t work. The tech is the impediment. I think that is really where everybody is playing.

Public acceptance, yeah that’s a factor, but again, not being on the sidewalk is a much smaller challenge. People are less pissed off at things that are not in their way. We’ve been trying to figure out how to be as good neighbors as we can, using the margin of the road, making sure that we’re not blocking traffic, other cyclists, those kind of things, but to be honest, that has not been the problem.

You just closed a $4.2 million seed round. Where do you expect to be a year from now?

Part of that money was to be deployed to Austin, and we’re rolling out soon, so that’s going to be pretty big for us. But I would like to be in hopefully another location by the end of another year. Most importantly, I think we would like to have even more vehicles on the road. So 25 is a lot for a company of our size, and I would like to add even more out there.

The other big thing is that we’re going to have our second generation vehicle by next year. That means cheaper, more reliable and easier to manufacture [vehicles] with the goal of really chipping away at trying to have a large delivery program in a couple cities in the U.S.

I remain optimistic and it’s clear to me that capital is not the limiting factor to making this work. If you look at Cruise or Nuro or Waymo, they have all the money in the world. If it were money, that problem would have been solved already. So it’s something else, which means that we have a hope, just like they do, of solving it. I’m optimistic that it is going to be scoping the problem and figuring out the right way of solving it as opposed to just throwing money at it, because that does not seem to have worked for the last five to eight years, at least in the AV industry.