The little robot follows Melonee Wise around a makeshift warehouse as she picks up boxes of cereal and packages of soap and drops them into a crate atop the machine. Freight, as Wise’s startup Fetch Robotics calls it, may be a machine, but its careful tracking of her movements recalls nothing so much as a dutiful dog.
The robot, which Wise demonstrated in a mock warehouse in a corner of the company’s San Jose headquarters, is one of two wheeled models introduced by Fetch in April as a way to automate warehouses and manufacturing buildings. While Freight is intended as an aid to human workers, the namesake Fetch has a single arm that can pick items off a shelf and drop them onto Freight, potentially replacing people.
Wise’s company is one of several robotics companies betting that robots, which have slowly found homes in auto plants and retail warehouses, are finally ready to roll out in much larger numbers. The CEO says in an interview I conducted for a recent profile of the young roboticist and entrepreneur that smaller and faster computers, improvements in artificial intelligence, and cheaper sensors are all combining to make robots cheaper (in Fetch's case, tens of thousands of dollars) and more capable.
Fetch is one of the most closely watched robotics startups thanks largely to Wise, a key contributor at the seminal robotics incubator Willow Garage, where she helped design and build several models, and a team of robotics veterans she has assembled. Fetch, which in June raised a $20 million round of funding from Softbank and previous investors Shasta Ventures and O’Reilly AlphaTech Ventures, has sold a few robots to pilot commercial customers. But Wise has bigger ambitions to create a platform on which software developers can create new applications. “They have a chance to create the backbone of autonomous robots,” says Shasta Ventures Managing Director Rob Coneybeer.
The blunt-speaking Wise, whose voice suggests a mellower version of the comedian Paula Poundstone, talked about how she got into robotics, what she hopes to accomplish at Fetch, how she aims to compete against Google and other companies snapping up robotics companies and talent, and the challenges of fulfilling her dream of a robot in every home. Following is an edited version of our conversation:
Q: Can you say where you’re doing the pilots?
A: No, we can't, but we’re doing three pilots with really big customers in warehousing and manufacturing. You have definitely heard of them.
Q: How did you decide to focus on that particular area, given that you've been trying all along to build for pretty broad application, even in the home?
A: At Willow, we spent two years trying to figure out what the next thing in robotics would be. The first year we tried to understand if there was any play in the home. The answer was a resounding no.
Q: Why?
A: The expectations are too high and the price tolerance is way too low. So people would love to have a robot that would do their dishes or tidy their house, but they want all of that for $500 or less. Even when you challenge that notion by saying, well, you know the Roomba you bought last month was $850, they're like, oh no, I bought that on sale.
There was this big hype about at-home telepresence. Everyone wants to put telepresence inside someone else's home, like their mother's, but no one actually wants it in their home. They don't like the privacy challenges.
Q: What’s attractive about logistics and manufacturing?
A: We strongly felt that logistics and materials handling and manufacturing was very scalable. There's a strong need for it. One of the things that sold me on it is there's a 600,000-person job gap right now for logistics and manufacturing. They just don't have enough people right now. Turnover is really bad. They also want to increase performance, and people have a rate limit. They get injured. There's shrinkage. When you pile all these things up, there's a great case for robots.
Q: To replace people?
A: We saw the need for not just pure autonomy but collaborative autonomy as well. That's why you see on our site two basic notions of how we might roll our product out. One is follow pick, where we have collaborative behaviors overlaid on top of autonomous behaviors. Then in the Fetch and Freight pairing of the robots, we have a pure autonomous behavior. It helps us cover more of the market.
There's definitely things that Fetch can't pick up right now--objects over six kilograms. People are still going to have to pick that up, but how can we make it easier for them to do that task, get more productive time out of them? Walking back and forth and doing transit is not productive. The robot can do that. Fetch also can't pick up deformable objects. Clothing is really hard for robots right now.
Q: What's the tradeoff between complete automation and flexibility?
A: Flexibility is really important for us. That's why we've targeted deploying in existing warehouses. Right now I don't think that 5 million-square-foot warehouses are the right place for us. At that point, you're probably going to be better off putting some conveyor systems in and then having robots locally in regions of the warehouse. We haven't tried a 5 million-square-foot warehouse, so I could be wrong, but right now we're targeting medium to large warehouses and environments.
We see the robots as platforms that have ever-increasing capabilities. We may roll the robots out with our base functionality today, but I can promise you in six months, it'll have a lot more capability than it had today. When you first get a Fetch, maybe the first capability it has is picking. Then we say, hey, we have this new capability for you, it can do light assembly, or kitting, or whatever's next.
Q: Will you do the manufacturing here?
A: Yeah, until we get to over a thousand robots, we're probably going to continue to scale internally. We can control the process a lot better.
Q: Are these built piece by piece, or have you built in some method to scale up production?
A: In our first run, we built 40 robots. The smaller robots take about four hours. We build seven of those at a time. The arms take the longest time--we build six at a time, which speeds up the process quite a bit. It's actually very fast. We designed the robot for quantities between zero and 10,000. If you want to go higher than that, you'd really have to retool the entire robot so you get higher quantities.
Q: Will $20 million last for awhile?
A: Yeah.
Q: But you've said it takes hundreds of millions to do a robotics company.
A: It definitely does. If you look at the companies that are coming out lately, on average they're taking between $50 million and $100 million to get to product, which is shocking. I don't know how much it'll take us. It took us $3 million to get to product. So we're ahead of the curve. But the growth stage will be the hardest for us, just because there's so much hiring to do.
It's a somewhat competitive market now. Certain companies have gone around the Bay Area gobbling up every roboticist in sight. That's going to be the biggest challenge for us. But we’ve got some really good pilot customers lined up and we're starting to do our pilots. We're shipping our first academic robots at the end of this week.
Q: How are those different?
A: The research edition doesn't ship with our commercial software, but the underlying drivers and base-level software all ship on that robot. It's for research and development. Private labs like a Samsung or Intel private lab, universities. The goal is to allow companies that kind of want to try it out to buy one. For researchers, they need a platform and it benefits us because long-term it helps us with hiring; a grad student comes out of college trained on our robot. That's a big win for us.
Q: What kinds of uses are being researched?
A: With the PR2 [Willow Garage’s robot], we saw some really crazy things that people did with the robots. I have no doubt we will see very similar crazy things. I remember the first time I saw someone putting their PR2 next to a live fire with a frying pan. They wanted the robot to cook an egg. I was like, whooooah. The PR2 covers were flammable. So it was really scary. One of the things we changed on our robot is all of the covers are fire-retardant.
Q: On the commercial side, what else might be next?
A: We've found that we can do all sorts of cool analytics in the warehouse. We built this tool to do basically a Wi-Fi survey of an entire building. You wouldn't think it's that interesting, but it's actually very interesting to a lot of customers. We can just say, "Robot, do a Wi-Fi survey of the entire warehouse." We can figure out where the dead spots are, where they have too many access points intersecting and causing issues.
There's a lot of things we're finding are cool utilities that we had to create for deploying it for ourselves but we're finding that the pilot customers get excited. We can provide Web applications for them to look at and administer. We're starting to do stuff like that.
Q: Are you still building a broader platform here, where there could be new and unforeseen applications, despite the relatively narrow focus for now?
A: I don't think there's any downside to having a platform long-term. It only enables our business to go further. Once we tackle the logistics market five or seven years from now, then we can go look at something else. I'm very optimistic but at the same time I'm very realistic about robots. It's a long, long, long, long, long process--I don't know if I put enough longs in there.
So many people think that robots have been around for so long. One of the first physical robots ever created wasn't created until 1954. Sci-fi got way ahead of robotics. When you show people Shakey, in 1970 crosses a room in eight hours and goes 15 feet, and people are like, Wait a minute, I thought we were already Lost in Space.
Q: Even in the DARPA Robotics Challenge this year, the stories were all on how robots were falling over.
A: And wheels won the day. Every critic of Willow would always ask, "Why did you build the PR2 with wheels? Why doesn't it have legs? Isn't that the stupidest thing in the world?" And we were like, “Well, but we don't want to spend the next 20 years working on walking.” It's still a long way off.
Q: Yet there's also the Boston Dynamics robots that can gallop and walk up hills. Are they not autonomous?
A: A lot of the Boston Dynamics stuff is carefully controlled. At the same time, it has four legs, which helps a lot. They've been working on that for a very, very long time and there's been a lot of money poured into that.
We're trying to do it cheaper. Casting Fetch’s arm all as one piece creates a larger structural stiffness because stiffness is so important, [but] our goal [also] was to try to make it lower-cost because there's far less machining. So by casting it, we've saved some time, some money, some assembly. People are actually very inexpensive, so you're constantly trying to chase being cost-equivalent to a person.
Q: What broader progress do you want Fetch to make, beyond these first markets?
Long-term we'd like to enable people to train the robots physically. Today, you could be building iPhones for six months, and the next thing you know you're building iPhone "n" for another six months. So when you have that kind of consumer product turnover, you need to be able to reprogram the robots effectively. So a lot of what we're focusing on is how to enable the customer to do more and more with their robots and make software as a service on top of the platform so they can expand the capabilities to their choosing.
Q: How far along are you on that?
A: We think it's achievable but it's definitely something we're working on over the next three to six months.
Q: How far back does your interest in robotics go?
A: I was always interested in building things. My dad was a technician, so I thought he was just the coolest person alive. He could solder and he could repair cameras and all these things, and I wanted to do that. So he got me a line-following robot, but I never wanted to follow any of the instructions, so I would just pick up a soldering iron and go for it. I did some stuff with it, but the problem with little follower robots, it's hard to be creative.
For me, Legos was the thing that changed that. When I was 7 or 8, my dad sent me to this camp at College of DuPage, a local community college near Chicago, where you got to play with Legos. The first thing I built was this plotter and you could draw with a pen in the holder. I struggled with it so much, but I thought it was so exciting. I was happy to be failing.
Q: What do you mean?
A: It wasn't like you do one thing and you're done. It's how do you come up with the solution to make it do something. It was really hard. I remember the first time I got to draw a smiley face, I was like, this is amazing. Then I was like, now how do I draw a cat? I escalated the complexity of the drawing to a factor of 10.
Q: And you stuck with that?
A: In high school, I did a lot of different things. I did photography. I did physics and math. I did all the AP classes but I was pretty much a slacker. My senior year, I took every Wednesday off until they caught me, playing video games, doing things I shouldn't have been doing. I almost didn't graduate.
My family was never very wealthy at all. I actually went to the University of Illinois on a Pell grant. I decided I wasn't going to make that much money as a photographer, so I became an engineer. When I got to U of I, I really wanted to learn everything in a very short amount of time, because I knew I couldn't afford to be there. I enrolled in both a physics degree and a mechanical engineering degree. I didn't attend class either because I had to work two or three jobs. Let me tell you, that's really hard.
Q: That’s when you got back into robotics?
A: I met Derek King, he's lead systems engineer here, and we started building robots together. We ended up becoming housemates, seven of us moved into this house on campus. No one wanted to be in the same room with us because we had robots everywhere.
Our first robot, Zippy, was a crying shame. We had proto-boards and duct tape. I was a mechanical engineer and the best I could do is tie a motor on to a piece of wood. We had no money. If we had used gum, it would have been too expensive. But because we didn't have money, it forced us to learn a lot of things. We had to design our own. We learned all these things because we didn't have money.
Q: How did you end up at Willow Garage?
A: I met with the guy who started Willow, Scott Hassan, and he offered us jobs. It was a funny conversation. Derek said, "Melonee, what are you doing, you don't know anything about this guy." I was like, "But it's robots!" So I took the job and came out here five days after I accepted it. Then Derek got hired at Willow. We've been building robots ever since.
Q: You've learned a lot of lessons over the years. Is failure the nature of robotics at least today?
A: Oh yeah. Some of the best roboticists are people who have worked on robots that have just sucked. Or had problems, or they tried to do something and it just failed miserably. I have failed so much, I don't know how many wins I have. The other night, I heard a great thing on the TV show Silicon Valley, the CEO of Hooli was saying, "Failure is just pre-greatness." That's exactly my life.
When we started designing Fetch, it was a total clean-slate design. The big thing that we knew we had to do is pick up heavier things. The only way to do that was to totally rethink how we were doing the arm design. The other thing we focused on was part count reduction. Fetch has about 500 unique parts. PlatformBot [another Willow robot] had 1,200 unique parts. There's a lot of things we did to address manufacturability. If it takes 45 minutes to take the computer out of the robot, it's a disaster. You can take the computer out of Fetch in about 15 minutes.
Q: Do you feel like you have a lot more trial-and-error to go, or have you reached some level of stability in design and engineering?
A: We've finally gotten to the point where we don't have to fear that we're going to be shut down tomorrow. The latest investment has given us the stability we need to take more risks. When you're afraid you're going to run out of money or you're not sure what the future holds or whether you have a runway, your ability to take risks is greatly diminished. Now, we have the ability to take those risks, and yeah, we'll probably fail at some of them, but some of them will really succeed and be so cool.
Q: What category of risks or challenges are next?
A: The big challenge/risk is the cost reduction part for the technology. Most of it's centered around the arm--stripping money out of the arm. Software is pretty easy to blow away and restart, the cost of doing it is relatively low. But with hardware, it's a huge capital cost to take a big technology risk. If it doesn't go well, you have nothing.
Q: You've talked about wanting to get a robot into every home. You're obviously not aiming for that here, but is that still a longer-term goal?
A: It would be really cool. I'm afraid at the rate robotics is going that it won't happen in my lifetime.
Q: Really? That long? That's a long time, at least for you.
A: When I think about what it means to have a robot in the home, I'm thinking of a mobile manipulation robot. I'm thinking Rosie [the robot maid in The Jetsons]. I'm probably one of the most pessimistic roboticists you'll ever meet. I constantly hear people say it'll only take three years to do X. When you look at autonomous cars, when the DARPA Grand Challenge finished, which was almost eight years ago now, they said, oh, in a couple years we'll have autonomous cars everywhere. I think we're still another good eight to 10 years out to practical deployment anywhere in the United States.
When I think about how long it's going to take us to deploy just into light manufacturing and how long it's going to take Savioke into hospitality, we're only at the first incremental step to getting into the home. I would love to see it. It's the labor of my life. It's something I'm going to work on probably for the rest of my life. But I don't think I'll feel bad if we only get halfway there, because we will have gotten halfway there.
Q: How did robot expectations get so far ahead of the reality?
A: It's a physical embodiment, so there’s a media fascination. Everyone says they want a robot to do their laundry, all I keep thinking is, build a different laundry machine. It's much more practical. Many of the things people want robots to do will become single-purpose machines.
Going into the home, you're not just deploying a robot, you're also trying to overcome hundreds of years of cultural behaviors and mannerisms. If you said to someone, "If you want a robot in your home, you need to change all your dishes to bright orange, you need to change all your cabinetry to be this color and it has to have these handles, and if you want a robot in your house, you have to have the robot house," I think you'd get some pushback on that. It's not your home, it's your robot's home, you just happen to live in it.
Q: Anything else on your mission here or generally?
A: Our big mission right now is hiring.
Q: You seem to like that sort of thing, unlike some technologists.
A: I didn't think at first I would like it that much, but it's actually kind of fun. I like seeing how excited employees and customers get about the company. Sometimes it can be very wearing. I've never considered myself a very extroverted person. The real test will be when my plate is so full that being in touch with the technology and robots is not something that I'm able to do. Then we'll see whether I really enjoy being the CEO or not.
