Interview: Kiwi Campus, a Colombian-Californian Startup, Is Rolling Out Autonomous Food Delivery in Berkeley

Kiwi Campus, a startup in Berkeley, California, has built its business model on using its advanced autonomous robots to deliver food to hungry college students.

Photo: Ricardo Rambal of Kiwi Campus alongside one of the startup’s “KiwiBots” in California. (Photo credit: Kiwi Campus)

The company was co-founded by Colombians and it maintains operators in Bogotá to monitor the robots in real time and troubleshoot any problems they run into — sometimes literally — in the field.

A sales team in Medellín is also supporting the pivot Kiwi Campus is starting to make from its current B-to-C concept to a B-to-B model.

To learn more about how the Colombian-Californian hybrid startup came to be, what the future holds, and how Colombian tech talent is making an impact in Silicon Valley, Finance Colombia Executive Editor Loren Moss recently sat down with Ricardo Rambal, part of the hardware team who works on manufacturing and maintenance of the robots for Kiwi Campus.

The delivery robots have been built to use machine learning to improve over time as they navigate the sidewalks of Berkeley, something that few other autonomous tech has attempted to crack. (Photo credit: Kiwi Campus)

Loren Moss: What is Kiwi Campus?

Ricardo Rambal: Kiwi Campus is a company based in Berkeley, California, that delivers food around the campus and to other places in a cheap and economic way.

Loren Moss: How does it do that?

Ricardo Rambal: What we do is that we have a system that combines people and robots to deliver food cheaper. We have a KiwiBot, which is a little box with wheels that rides by itself around the Berkeley campus, and it’s aided by some people to upload the food and then the robot goes to the destination.

“What Uber and Tesla have been doing is trying to be autonomous on the streets beside other cars. What we do is we deliver on the sidewalk, which could be cheaper and faster.” – Ricardo Rambal of Kiwi Campus

The client orders food, the KiwiBot goes to the restaurant, and there it meets someone who uploads the food. The KiwiBot is a little box with a lid. So, by uploading the food, I mean it actually opens with a motor and then the people put the food inside and it closes again. After that, the KiwiBot knows where to go and goes to your house to deliver the food. When it meets you downstairs, the KiwiBot opens again the lid and you can take your food out of there.

The important part about this is that this is cheaper than actually hiring people, mainly because, with one person on the field, we can oversee more deliveries. While a KiwiBot can run by itself, one person can deploy five or seven and oversee them.

Loren Moss: How does it get around obstacles?

Ricardo Rambal: The KiwiBot rides around the sidewalks. It rides semi-autonomously. By this we mean that it recognizes its context and it accelerates or stops depending on the context. If there is somebody walking nearby, it recognizes and slows down. If there is plenty of room, it keeps on going and that is pretty much it.

Loren Moss: And when the student or person who made the order comes downstairs or outside, the KiwiBot is waiting for them?

Ricardo Rambal: Yes. For now we deliver to the doorstep of the main floor where you ask the delivery to be done. When the KiwiBot arrives, it tells the customer to open the lid. So the customer knows which bot is attached to their service this way, and also by an order number and by the application on their cell phone.

Loren Moss: It sounds like you are using artificial intelligence (AI) and machine learning technologies to recognize the surroundings?

Ricardo Rambal: Yeah. The KiwiBot uses lots of artificial intelligence technologies and we do use machine learning — and specific programs of machine learning like deep learning — in order to use data for the robot to improve its techniques. We also use computer vision, which is capable of understanding context from a single picture and we use basic, old-school control using electronics.

Loren Moss: How difficult has it been to get the licensing and any other type of permissions required from the University of California, Berkeley, and also any other agencies that may have regulations? Perhaps the FCC for the remote, radio control aspect?

“We are running a pilot with DoorDash where the ‘dasher’ is not a person anymore but a KiwiBot. So we are aiming to do that more so that we can focus more on the technology rather than this old, other-world problem of logistics by itself.” – Richardo Rambal

Ricardo Rambal: The first answer is that, thankfully, we were part of — and are still connected to — SkyDeck, which is one of the University of Berkeley accelerators programs. So by being part of the accelerator, we get the permission from the UC system quite easily.

As for control systems, our technology is based in 4G LTE, so that means that each robot is basically a cell phone. What we do is contract the data system from a third party, and we don’t have to ask for permissions to use the radio frequencies.

Loren Moss: How did you come up with the idea of this product?

Ricardo Rambal: Felipe Chávez, our CEO, ran a successful business in Colombia called Lulo, a delivery company with humans using an app. That was about three or four years ago. That company was sold to another delivery company in Colombia, and his idea after that was to go to the universities. In Colombia’s case most of them are closed and the campuses are big.

His idea was to deliver inside the university with students. You would hire some students and, in their free time, they could make some extra money by delivering goods from the outside the campus to the university.

That model worked for a while, but he wanted more. So he went and he tried to replicate this model, which was successful not only in Colombia but in Chile, in the United States. But then he found out it was a very different context and the students, at least in Berkeley, weren’t willing to work part-time jobs. And if they were, they would expect really high pay. There is no surprise in that. The college’s tuition is high.

Loren Moss. And California is one of the most expensive states in the United States.

Ricardo Rambal: Right, costs in California are high. So in order to keep on going with the delivery services for students, he had to pivot and he had this crazy idea of delivering with robots — of delivering with boxes on wheels.

That’s pretty much how we got started, and it started with a small prototype. It was an Ikea kind of box with a cell phone attached and a small radio controlled car on the bottom. He ran the first test, he felt he could do it, and he started moving really really fast. We have 50 robots that were made in China and designed in California by Colombians. The robots of China were the last fleet, and currently, the model is operating, and we are making profit.

Loren Moss: How did you obtain funding to launch at the beginning in the development stages?

Ricardo Rambal: We had some money that Felipe got when selling the other company. But mainly, because of the SkyDeck accelerator program and also some other investors from Italy that are in food startups.

Felipe is a great salesman. He is a really good CEO. As the company was getting bigger, we could show more to the possible investors. So what has kept us alive in the startup life is to be quick. A key value of any startup is to iterate fast — not only on the hardware but in software. We have many more lines of code every day being uploaded to the main system.

Loren Moss: How big is the team?

Ricardo Rambal: The team is divided into three parts. One is Bogotá, where we have the supervisor for the autonomous riding. Another part is in Medellín, which is a little bit of sales and also some engineers. And we’ve got the big part, which is operations and engineering in Berkeley, California. We are talking about 30 people in the team.

Loren Moss: You said sales out of Medellin. Who is it that you sell to?

Ricardo Rambal: In the last few months, we are trying to change from a B-to-C kind of business to a B-to-B kind of business. We found out that we could make much more — not only profit, but we have much more volume in the technology we are developing in the actual logistic system by itself. So what we are doing is that we are selling the service of having a robot to other delivery companies.

For example, there is a big one in the Bay Area called DoorDash, and we are running a pilot with DoorDash where the “dasher” is not a person anymore but a KiwiBot. So we are aiming to do that more so that we can focus more on the technology rather than this old, other-world problem of logistics by itself.

Loren Moss: In Bogotá, you have supervisors? What do they do?

Ricardo Rambal: By the law of the state of California, every single autonomous driving vehicle has to be overseen. If the robot loses control, someone from Bogotá is ready to take over so that it doesn’t cause an accident. They take over and they’ll wait for it to be replaced. Or, if they can fulfill the delivery, they’ll just do it by driving the robot from Bogotá.

Loren Moss. How do you plan to defend yourself against the giants that are taking over the space? Uber Eats, for example, is already here in Colombia. They use people, but we know that Uber is already experimenting with autonomous technology. And there are other companies out there. Amazon has experimented with drones. These are giants — and there are also smaller companies that are doing similar things. How do you differentiate yourself, and how do you armor yourself against other competitors in rapidly moving market?

Ricardo Rambal: The good thing about the startup context is when you are a startup, those big companies are willing to buy you. In the startup life, either you get big or you get bought. Those are the two paths that you can take, and either of them is good.

Loren Moss: Or you can get bankrupt.

Ricardo Rambal: That’s the path that you don’t want to go, right? But Silicon Valley and the Bay Area is such a good environment for this kind of stuff that we don’t see the big ones as a threat but as something that can purchase us or help us grow.

“In the startup life, either you get big or you get bought. Those are the two paths that you can take, and either of them is good.” – Ricardo Rambal

For example: DoorDash. It’s in the delivery market, but they were not into robots. We are in the delivery market, but we make robots. We are not killing each other. We are actually working together.

But it’s also because we have one big differentiator: We drive autonomously on sidewalks. What Uber and Tesla have been doing is trying to be autonomous on the streets beside other cars. What we do is we deliver on the sidewalk, which could be cheaper and faster.

Loren Moss. Why not deliver with drones?

Ricardo Rambal: Well, I personally believe that drones are amazing, and I love the technology. But drones they have two main characteristics that make them not a good solution for last-mile delivery.

First of all, humans are not used to something big being on top of them. For example, imagine the old classic cartoon of the piano lifting in the air. It’s a stressful situation for humans to have objects above their heads. Then there are regulations for drones flying in the states. For example, you cannot fly drones in national parks if you don’t have a special permission. You cannot fly them in the city if you don’t have the special permission.

So, drones are amazing, but they can be really hard to control. And when they are easy to control, there is a second problem: battery life. For lifting things and flying you have about 15 minutes of run time, and that’s too little time to make a successful delivery in terms of going and coming back. Our robots have batteries that can last six hours of run time. That’s a huge difference between them that makes the robots way cheaper.

Loren Moss: What’s the next step? What is your company going to look like two years from now?

Ricardo Rambal: We are hoping that we get many more customers and we grow wider to different campuses. Or we might even go fully into a city.

I hope that in the next two years we have the technologies settled and stabilized so that we can have more robots and less people, and then we can have more teams in different cities of the states.

I don’t see it happening in Colombia, at least for the next two years at least, but I believe in the next 10 years it has to happen here as well.