Starting out as iRobot's first official employee, Joe Jones has had a remarkable career in robotics. He spoke with TechRepublic about deep learning, the history of iRobot, and the power of small robots.
In 1988, when Joe Jones was researching small robots at the MIT AI Lab (now CSAIL), he had an interesting idea. "I got excited about the really small, reactive robots, and thought: you can do a lot with this. You could build a robot that could clean your floors."
Jones began constructing LEGO versions of a floor-cleaning robot he called the "Rug Warrior," and entered it into the Robot Olympics.
Three years later, when he started working at Denning Mobile Robotics, he brought the idea with him. Together with his colleague Jack Shimek, a mechanical engineer, Jones developed the proof of concept. Then they demonstrated their idea of RoboBroom to the whole company. "We thought: 'This is going to take off!'" he said.
Ten days later, they were both fired.
"They thought that this was going nowhere, that we were playing with toys," said Jones, "that real robots were big and expensive."
Jones didn't let this setback deter him. A few months later, in February 1992, he was hired by iRobot, a new company that had started when Colin Angle, (one of the three founders), received a $20,000 contract from a Japanese company to build small robots. Jones was the first full-time employee. When he left, almost 15 years later, there were 300 employees—and at the peak, there were nearly 600.
But in the early days of iRobot, the company struggled financially. "Colin used to joke that in the first two years," said Jones, "they never had enough money at the beginning of the month to make payroll at the end of the month."
"And, I'll add—sometimes they didn't have enough at the end either!"
So while his colleagues at iRobot were receptive to the idea of a floor-cleaning robot, it wasn't until 1999 that the company had enough money to develop it—at that point, the number of employees was up to 20. Jones and his colleague Paul Sandin worked on the Roomba full-time for the next three years, and the final product launched in 2002.
"This time," said Jones, "We weren't laid off."
Roomba became one of first, popular, home robots, and has since sold more than ten million units across the globe. Despite the growth of the company, Jones was restless. "The company had been successful with the Roomba, and, shortly after, a military robot called PackBot," said Jones. So iRobot continued to put money into developing small robots. "But I just didn't think the next big thing in robots was in those areas, he said.
"One of the problems with Roomba is that after you develop a robot that cleans the floor, what do you do next? Clean windows? Clean the toilet?" he asked. "The technology to do that has absolutely nothing to do with cleaning the floor—you're starting from scratch. The only way I could pursue what I wanted was to leave the company."
In 2006, Jones convinced a few others to leave iRobot to start another company, Harvest Automation, where they built a robot for the greenhouse industry. "In agriculture," Jones said, "the technology is related. If I build a robot to do one thing, I can build one to do a similar task."
But while Harvest started out doing well, shipping to 30 farms across the country, the market developed slowly. The company decided to branch out into warehousing. "In 2012, Amazon bought the warehouse robot maker Kiva Systems for 775 million," said Jones. "It proved there's a billion dollar market for robots in warehousing, so now, a number of people are jumping into that." Harvest's new warehouse robot, expected to be on sale early next year, will compete directly with Kiva.
Eight years after he started Harvest, Jones decided he wanted a new challenge. He co-founded Franklin Robotics, where he continues to work with small machines, this time with the goal of selling small robots to gardeners.
In his own words...
Deep learning has become popular in AI lately. What's the history behind it?
This started back in the '80s, when they realized one level deep didn't work.
At some point, people thought, what if there are several levels of neurons? It's what they call neural networks. The thing I wondered was, "how many levels deep are people?" Every time you add a level, there's a delay. You know how fast people are to respond to people. But I didn't like using neural networks. If the system incorrectly identified a photo, there was nothing you could do except show more pictures and see if it got it right. But you couldn't tell what was going on inside the system. So, nobody could figure it out and it fell out of favor. More recently, a few people, like Jann LeCun, kept working on it, and now it's much better. There were all kinds of other techniques in between, a lot involving search and reasoning, but now deep learning has beat them all.
What's the relationship between drones and robots in agriculture?
Most of what drones do is gather information about crops and growing conditions. Any information you can get by looking from above. But if you need to take action from the ground, or look at the base of a plant under a canopy, drones can't help with that. That's why you need ground-based robots.
What's the biggest area of need for robots today?
There's a huge need for robots in agriculture. The UN report, "How to Feed the World in 2050" shows that we will have to increase agriculture production by 70% to feed all the new people on the planet—and this is when you can't use any more land or any more water. Agriculture has to get a lot more efficient.
Harvest is moving towards warehousing. which is the right decision for the company. You have to build a gigantic software system that runs on a central computer that tells robots where to go—but I wasn't interested. So I stopped working there. At Harvest, we sold big products to big agriculture. I started thinking—what if we sell small robots to gardeners? Ultimately, the goal is the same thing: to help out agriculture. In a decade or so, robots will grow our food.
- Why robots still need us: David A. Mindell debunks theory of complete autonomy (TechRepublic)
- Smart machines are about to run the world: Here's how to prepare (TechRepublic)
- Chinese factory replaces 90% of humans with robots, production soars (TechRepublic)
- Why AI could destroy more jobs than it creates, and how to save them (TechRepublic)