This Woman Makes Robots. And No One Is Going to Stop Her.
Melonee Wise was employee 2 at Willow Garage. Its legacy destroyed her first startup but lives on in her latest venture.
--
Melonee Wise paces through the aisles of a small mock warehouse, pausing every so often to take an item from a shelf and drop it into a blue bin beside her. When she moves on, the plastic bin follows, like an obedient pet.
The bin sits on top of a wheeled robot called Freight. Wise has been developing Freight and its sibling Fetch, a torso with one arm, for the last six months. She is one of the most watched and most admired roboticists in the industry today, and these creations are the latest embodiment of her worldview, one that was cultivated first at the famed Bay Area laboratory Willow Garage, followed by a short-lived startup called Unbounded Robotics and now Fetch Robotics, where she is CEO.
Wise walks down the aisles, as humans do, pausing and changing her path as new whims arise. Freight reacts flawlessly, following on her heels. The human and the robot are working together in casual harmony, bringing to life a dream many have nurtured for decades.
But this collaboration is not just an idealistic pax de deux to promote interaction between flesh and metal. Fetch and Freight exist so that e-commerce companies big and small can afford to automate their warehouses. Unlike the robots in Amazon warehouses, which require specially made shelving and an actual human to pull goods off of a shelf, Fetch and Freight can drop into nearly any stockroom and autonomously ferry items from place to place. Imagine an e-commerce business where human hands never have to touch the goods.
Wise has been trying to bring about this vision for years. Her earlier startup, Unbounded Robotics, shut down in August 2014, just as it was about to ship its first robot, UBR-1. Yet this spring, just a couple months after announcing Fetch Robotics and $3 million in Series A funding, the core team from Unbounded was already showing off two mature robots. Fetch is clearly a blood relative of UBR-1.
But to really understand Fetch’s pedigree, you have to understand Wise’s.
As a child, Wise wanted to be a photographer; maybe the next Ansel Adams. Her father, a Polaroid technician, brought home broken cameras and fixed them up for her. She liked capturing black and white shots of architecture and everyday life and developing them by hand.
In her grandmother’s Montessori daycare she was prodded to keep trying new things, and she found that she loved math and building with blocks. She especially loved her massive collection of Legos.
But after Wise enrolled at the University of Illinois Urbana-Champaign in 2000, she picked up a hodgepodge of engineering majors and minors. She had grown up poor, and pursuing photography as a career didn’t feel like a smart way to start making money.
Still, she believed her work should be fulfilling. She spent her summers interning at big firms, such as Honeywell, Alcoa and Chrysler. The engineering jobs felt small, as if she might spend her whole career never building anything interesting.
“I felt like I needed to get into an environment that could embrace me,” Wise says. “I could be myself and I didn’t have to wear stupid khaki pants and a polo. I just didn’t want to settle.”
She considered becoming an astronaut, but balked at the competitive selection process. The place she felt most at home was in school working on robots. She had befriended a close friend’s roommate — Derek King, a quiet guy who lit up when one of his interests came up in conversation. They both liked ultimate frisbee and building robots in their spare time. In 2003, they collaborated on their first one: Zippy, an autonomous bot that wasn’t much more than a few motors tied to a piece of plywood.
The “sad engineering reality” of a boring job felt inevitable, but Wise wanted to put it off. She was accepted into the mechanical engineering PhD program at the University of Illinois and celebrated with a tattoo on her wrist of a turtle, representing the slow road to graduation. She lasted three semesters.
During the first year of the PhD program, Wise and King joined a team building an autonomous car for the DARPA Urban Challenge. It was exactly the kind of opportunity they had always wanted; cutting edge robotics and the chance to work with a network of University of Illinois alumni in the industry. A year later, in December 2007, they flew to California for a crucial demonstration to DARPA officials.
While working on the car they met a man named Scott Hassan, who provided the four Ford Escapes the team was hacking into autonomous vehicles. Hassan, who had helped code the original Google search engine, had made a bundle from that and from subsequently selling a startup to Yahoo. Now he was turning his wealth toward the robotics industry. He offered Wise and King jobs at his new laboratory, Willow Garage.
The night before the big DARPA site visit, the car’s electronics fried. The project was over. King went back to school, but Wise, bored with her PhD, accepted the Willow Garage position.
“I just took a leap of faith,” Wise says. “I would never guess that the company I became the second employee of would become one of the most influential companies in robotics of the decade. It was just pure dumb luck.”
The first years at Willow Garage were frenzied. Wise’s office mate slept at the lab so he could spend just a few more hours on his work each day. Teams built a robot in a matter of months, and then turned around and built another.
“It was so great to be surrounded by all those people … who ended up being big names in robotics,” Wise says. “I can’t even describe the excitement that everyone had. It felt like we were changing robotics.”
Willow Garage set out to solve the problems that had held the industry back for decades. At research institutions, teams tended to spend years building a robot’s basic hardware and software. An extraordinary amount of money was poured into creating a machine that could do only a few things. Wise calls them “snowflakes.”
The solution, which Wise believes in to this day, was to create platforms. A company or research center can start with a versatile, pre-built robot or software and jump right into building applications for it. Willow Garage’s team built ROS, an open source operating system, and a robot called PR-2. Wise was on the team that first started imagining applications for PR-2 — everything from fetching a beer to plugging itself into a wall.
While PR-2 units trickled out, ROS exploded. Wise had written a deep well of tutorials that made the software unusually accessible. She also founded and ran a large internship program. At the end of the summer, the interns returned to their universities and spread the operating system like a virus. ROS is still the brain of diverse robots all over the world.
Wise then went on to build TurtleBot: a short, cylindrical robot on wheels that uses a Kinect to see its surroundings. Institutions can buy dozens of $1,000 TurtleBots for the price of a single, more complex robot. Willow Garage, and now the Open Source Robotics Foundation, have sold thousands.
“Willow had a big impact, and Melonee had a big impact on Willow,” says Steve Cousins, who was CEO of Willow Garage from 2007 to 2013. “(Hassan) saw she was going places. She’s got the eye of the tiger.”
Yet Willow Garage still wasn’t commercially viable, and the team spent the next year and a half searching for a way forward. Wise moved into a management role as the lab considered seven different industries. It never found a clear next move and began spinning off small sets of employees as profit-oriented startups.
In January 2012, Wise gathered a team at Willow Garage to build a bot that could roll around a room and move objects — what’s known as a mobile manipulation robot. They had instructions to incorporate a low-cost robotics arm developed by another group at the lab. They called it PlatformBot.
Wise’s team worked insane hours to make it happen. King had been hired on a few years back, and their duo expanded to four as they grew close to mechanical engineer Eric Diehr and software engineer Michael Ferguson.
It was all for naught. The arm didn’t work well, and the team made a series of bad decisions in the r0b0t’s design.
They started talking about spinning off a company so they could start over. But just as their plans were forming, news broke that Willow Garage had six months until it shut down. Hassan, who had been funding the lab with his own money, was no longer able to sustain the staff. Wise, King, Diehr and Ferguson broke off in February 2013 and called themselves Unbounded Robotics, trusting that they could develop a concrete spinoff agreement once Willow Garage’s situation settled down.
At Unbounded, they developed and launched their own mobile manipulator within eight months.
“We fundamentally started everything over from a blank slate,” Wise says. “The stuff that we had learned at Willow, it was obvious that we had gone down a bad path. It just wasn’t going to work.”
This time it worked. The one-armed UBR-1 was slated to retail for $50,000 — one eighth the cost of Willow Garage’s $400,000 PR-2.
The team finally sat down with Willow Garage in December 2013 to write a spinoff agreement. The final document was disastrous for Unbounded. Although UBR-1 didn’t incorporate any of Willow Garage’s patents, Unbounded did have permission to use the IP. It wasn’t crystal clear that UBR-1 belonged to Wise and her team, and that made investors wary enough that Unbounded couldn’t raise its first round of funding. They decided to abandon the company.
“That was a really tragic time for all of us,” Wise says. “We were all very lost.”
Not long after Unbounded shut down, the team saw a vague job posting seeking “robotics gurus.” Steve Hogan, of technology incubator Tech-Rx, wanted to get in on the coming robotics boom. He hired Wise, King, Diehr and Ferguson. The opportunity felt perfect, like a free pass to jump back onto the road they had started down so many years prior.
They spent six weeks researching markets such as robotic arms, hospitality and elder care before settling on e-commerce as the target for their future robot. The bulk of their expertise now lay in mobile manipulators, and the space was still fairly empty. The path was clear to create a product that would give a small company the robotic power of an Amazon fulfillment center.
Fetch Robotics launched officially in February with funding from O’Reilly AlphaTech Ventures and Shasta Ventures. “(Wise’s) name came up several times when looking into the robotics industry,” Shasta Ventures managing director Rob Coneybeer says. “We think it’s one of the best robotics teams in the world.”
To finally capture their Series A funding, Fetch had to leave behind all of the Unbounded IP and, once again, start over. Which was not a bad thing. “It’s actually pretty liberating to be forced to think in a new direction,” Wise says. “I think sometimes as an engineer you cling to what you know and what you’re safe with doing. It totally changed how we approached the problem and made us more creative.”
In an empty back room at Fetch Robotics headquarters in San Jose, a Fetch robot moves each axis of its long arm in turn. It raises and lowers its spine, growing from the height of a child to that of an adult. It will run 24/7 until it dies so the team can spot and fix its weaknesses.
The Fetch’s tongue depressor-shaped torso, arm and head are bolted on top of a Freight, making it mobile. Fetch can roll into a warehouse and start picking items off of shelves and dropping them into a second Freight, this one carrying a bin and traveling at its side. A new Freight, each of which can carry up to 110 pounds, can swap itself in after a full one rolls away — no human input necessary. The upfront cost of one Fetch is the same as hiring a human employee for a year.
Both Freight and Fetch are designed to be as low maintenance as possible. After being trained on the layout of a warehouse, there is very little they need to be taught. The robots automatically dock when they need to be charged. They resist being shoved by humans, or hurting them.
Wise invites a Freight to follow her around the office’s dummy warehouse by tapping a button on her phone. It uses a beacon sent from her phone, but it also sees every step she takes through a set of sensors that line its front. When a shelf or person blocks its path, it waits and then makes an arc around the obstacle.
Wise fills the Freight bin to the brim with Pringles, soap and other knick-knacks that the Fetch employees had picked up at a local dollar store (the cashier asked if they were preparing for the end of the world). Then she taps her phone and the Freight turns and rolls out of the room. In a real warehouse it would ferry its bin to a shipping area.
Wise’s role as the picker could be replaced entirely by a Fetch, which uses sensors in its head to locate and grab items with its pincer-like hand. A warehouse’s database integrates with Fetch’s brain to let the robot know the area in which to find an item, but it is up to the robot to pick out an individual object.
I had visited the Unbounded team, and when I look at Fetch, I feel a sense of déjà vu. Its form is noticeably similar to the UBR-1 robot Wise developed at Unbounded. But that is inevitable, according to Wise. All mobile manipulators end up with a base, torso, arm and head. If UBR-1 and Fetch are siblings, so are their competitors.
But take a look inside, and you can see what is different. Where UBR-1 had 1,000 parts, Fetch has 500. The arm moves via an entirely different mechanism. Its vision is better. Wise believes the company has already developed significant new IP, and will continue to discover more.
“So much of what we have to do now is so different from what we thought we were doing then,” Wise says. “Obviously Fetch and Freight are an evolution of all of our experiences.”
Like UBR-1 and PlatformBot, Fetch is a platform. This first model is built specifically for picking in warehouses, but there is nothing stopping developers from giving it other tasks. Fetch could stock shelves or pack boxes. Eventually, it could work somewhere outside a warehouse.
Packaging Fetch and Freight together with warehouse-ready abilities is a shift in marketing strategy for Wise’s team, and a smart one. UBR-1 was also capable of stocking and picking, but that use was only one among the huge array of applications for which the robot was built. Concrete abilities, not possibilities, are what could place a Fetch Robotics fleet in every warehouse in America.
Fetch Robotics is working with unnamed partners to test the robots in real warehouses over the next few months. Companies have already pre-ordered units ahead of the commercial ship date later this year. Wise won’t set the pricing until after the pilot tests, but she says a robot working 8 hours a day would pay for itself in a year. Ordering multiple robots will drop the price even further.
A lot of things have become clear to Wise since she began building commercial robots — things unimagined by the young woman who eschewed big corporations because she did not want to settle. She’s no longer naive enough to sit on a spinoff agreement. She’s hiring a large team, with a total of six former Willow Garage staffers now at its core. She’s mulling spaces beyond the warehouse for her future robots. After so many years, she believes that this is the robot that will make it.
“My life goal and ambition is not to become the best person in robotics and be known for it,” Wise says. “It’s to build something that works.”