The future is not yet now | March 15, 2013 | Palo Alto Weekly | Palo Alto Online |

Palo Alto Weekly

Cover Story - March 15, 2013

The future is not yet now

Local robotics companies are pushing the boundaries, but the Jetsons' Rosie is still years away

by Bryce Druzin

With a rotating head and two strong arms, the Personal Robot 2 from Menlo Park-based Willow Garage can navigate by itself and pick items up off the floor, open a refrigerator, fetch a beer and fold towels.

But before anyone gets too excited about the prospect of a personal robot maid — think Rosie from the '60s cartoon show "The Jetsons" — a few things: It costs $400,000, weighs almost 500 pounds and can't move up stairs.

And the towel it can fold? That'll take a couple minutes.

"One of the issues we have in robotics is that people's expectations of robots are really, really high because of what they see in Hollywood," Willow Garage CEO Steve Cousins said during a January presentation of the robot at the Avenidas senior center in Palo Alto.

An affordable robot with human capabilities isn't a reality yet. But from the humanoid, general purpose PR2 to simpler robots that perform single tasks, researchers and companies in and around Palo Alto are working to make robots ever more useful and practical.

"Fifteen years is all the time we've had cell phones, but yet we take them for granted," Cousins said. "So the world is changing fast, and I think the same thing's going to happen with robotics."

Willow Garage was founded in late 2006 by Scott Hassan, founder of eGroups, a group email-messaging company that would become Yahoo Groups. Hassan founded Willow Garage to "accelerate the development of non-military robotics and advance open-source robotics software," according to the company's website.

The first PR2 was rolled out in 2008. Today there are nearly 50 of the robots, nearly all of them at academic institutions, including 11 loaned, free of charge, to research institutes in the United States, Europe and Japan.

The robotics lab at University of California, Berkeley, under the direction of Pieter Abbeel, was one of these recipients. They generated a bit of a stir when their robot successfully folded 50 towels in succession, taking around 24 minutes per towel. Since the PR2's first towel fold in 2010, they've been able to get the time down to two minutes.

More recently, in 2012, a Berkeley research team collaborated with Google to program the PR2 to de-clutter a room.

Using the Google Goggles image recognition application, the robot was able to identify objects such as a mustard bottle and bar of soap, grasp those objects, and move them to appropriate areas.

Professor Ken Goldberg, a member of the research team, believes a commercially viable robot that picks up around the house could be a reality in the not-too-distant future. He thinks the object recognition and grasping programming to make such a robot practical can be developed within five years. He said the key would be creating a physical robot that's reliable and affordable.

"For under $5,000, people are going to start wanting it," he said.

Like Cousins, Goldberg also thinks robotics are close to a "tipping point."

"When you get capability it drives acceptance, which drives down the price, and drives further acceptance," he said. "So you get this positive spiral."

In this experiment, the PR2 used the concept of "cloud robotics." Instead of storing data for image recognition and grasping motions on hard drives located on the PR2, this information was stored on a remote network that the robot accessed wirelessly.

By outsourcing data storage and computing power to external networks, robots can take advantage of increased computing power, reduce the amount of physical storage they have on them, and share information with each other.

Both the PR2 and the Robot Operating System software it uses have their roots at Stanford University.

The PR2 is named after the PR1, which was developed at Salisbury BioRobotics Lab at Stanford. Doctoral students Keenan Wyrobek and Eric Berger, in collaboration with professor Kenneth Salisbury, designed and built the robot in a span of 18 months.

The PR1 was able to tidy up a room, unload a dishwasher and fetch a beer from the fridge and open it, though there was a catch: The tasks were performed while the robot was being controlled by a person.

Still, Salisbury said, he had long been interested in building a robot that was more than "just a hand on an arm."

Salisbury said he favors wheeled robots over legged robots because the challenge of balancing on legs still presents a difficult problem.

"I'd like to do something today that involves the perception and manipulation of objects, without having to worry about 'Am I going to fall over?'" he said.

At the same time the PR1 was being developed, Stanford's Artificial Intelligence Laboratory, under the direction of Andrew Ng, was developing an operating system for robots to use.

Salisbury said the PR1 served not only as a platform to study robotic capabilities but also as a tool to raise funds for the next generation of personal robotics.

"The founder of Willow Garage saw what we were doing, saw what Andrew Ng was doing, looking at both the hardware and the software capabilities that we were developing, and became quite excited about it," he said.

Wyrobek and Berger were hired by Willow Garage to head up its personal robot project, which produced the PR2. And the software originally developed at Stanford's AI Lab would be developed further by Willow Garage and become known as the Robot Operating System, or ROS.

ROS is now the most widely used open-source robotics software platform. Because it's open-source, the code can be freely viewed by anyone.

During his January talk in Palo Alto, Cousins said the open nature of ROS was bringing robotics closer to "exponential growth."

"Suddenly other people can look at what you did, learn from it, build on it, and then do something else and contribute back," he said.

Stanford was also one of the 11 institutions to be loaned a PR2 by Willow Garage in 2010.

Salisbury said one of his many dreams in robotics was to have a robot fetch him a cup of coffee. A simple task for a human being but not so easy for an autonomous robot.

"That's a pretty hard problem if you think about it," he said. "How do I get from here through the doors, through the elevator, somehow communicate with the person serving coffee, take the coffee (and) bring it back down?"

Last summer, student Tony Pratkanis was able to program the robot to perform an approximation of the task. The PR2 used laser sensors to detect obstacles, and it used accelerometers and its vision to determine how many floors it traveled.

Communication with the barista was aided by a handwritten note, and an enclosed container of juice was used in lieu of coffee to avoid spillage. All in all, the PR2's journey took 40 minutes for something that would take a human around 10.

"The coffee-fetching robot is a pretty sophisticated example of full autonomy," Salisbury said. "Yet that is a laboratory demonstration. That robot's not going to be serving coffee for (a few) years."

While Willow Garage and academicians work on creating a general purpose robot that can be programmed to perform multiple tasks, other people are working on simpler, single-task robots.

But if a robot can only perform one task, what's the difference between it and an appliance, like a toaster?

FoldiMate CEO Gal Rozov is developing a laundry-folding robot. He said a robot is characterized by how it responds to changes in its environment.

"If you are aware and do not do anything, you are a sensor," he said. "If you are aware and you react, then you can start to be considered a robot."

Though Rozov said he was inspired by watching "The Jetsons" on TV as a kid, he's setting his sights on something a little less ambitious than a humanoid robot like Rosie.

"You want to find a robot or a solution that will solve a specific pain point," he said, "And if we look at it like that, the time to market will be a lot faster."

Rozov identified folding clothes as a pain point.

"It's a tedious task," he said.

He started work on the concept in 2010 and founded FoldiMate, based in San Mateo, in 2012. A few months later, he was joined by Robert Ford as well as Ted Selker, who designed IBM's in-keyboard pointing device Trackpoint.

Rozov, who has a software background and is writing code for the FoldiMate, said he's working to have a functional prototype ready by April. Users will place an article of clothing on top of a rectangular box about 2-feet wide and 2.5-feet long. A platform will lower down into the machine, where the folding will take place.

Users will indicate whether the garment is a shirt, towel or pair of pants. FoldiMate will sense its size, thickness of fabric and whether it is long-sleeve or short-sleeve and adjust accordingly.

The company decided in early March to start referring to FoldiMate as a "smart appliance," as opposed to a "robot," in order to make it more relatable to customers.

"If you say 'robot,' it's the future," he said. "When you say 'smart appliance,' it's right now."

Rozov said he's first targeting laundromats and wants to sell them a coin-operated model for their customers to use. He hopes to eventually sell a home model that would cost under $800.

A robot that is a bit more autonomous than the FoldiMate is made by Neato Robotics, based in Newark and founded by Stanford Graduate School of Business students.

The company makes robotic vacuum cleaners, the first of which were sold in 2010. Nancy Nunziati, vice president of marketing, said the company was founded in 2004 and benefited from coming out after the Roomba, which was first sold in 2002 by Massachusetts-based iRobot.

"When (Neato) was founded, there was a lot more fear about calling things 'robots,'" she said. "There was a concern that people would react negatively or not be ready for robots in the home. And I think that was true at the time, but I think it has changed quickly."

Eric Ivers, general manager of RoboteX in Palo Alto, said he's also noticed a change in how the public reacts to robots.

His company makes security robots for law enforcement and other emergency responders that can be outfitted with cameras, manipulator arms and gas and radiation detectors.

He said that when he used to drive the robot around Palo Alto streets, people would stop and ask questions. But about a year ago that changed.

"It was just sort of 'Yeah, there goes another robot,'" he said.

RoboteX's robots are controlled by a person using a remote and watching a screen that shows what the robot's cameras see. SWAT teams are some of the company's biggest customers, and they've used the machines to check out houses where they suspect an armed person might be present.

Ivers said that while some definitions of robots include autonomy as a necessary feature, in the security industry "if it can be operated non-line-of-sight, it's pretty much considered a robot."

RoboteX is planning on releasing a model for home use in June that Ivers described as a "very stripped down version of our basic robot."

Users will attach a device such as an iPad or iPhone that would act as the eyes and ears for the user, who would control the robot remotely from the Internet.

The home robots look much tamer than the security model, which Ivers said was a deliberate choice. They're also blue.

"We didn't make a big, black, scary-looking robot for home use," he said.

Professor Salisbury at Stanford echoed the sentiment that if a robot is in a domestic setting, it should look pleasant. He also said it's important that domestic robots act in a way to reassure people that they are friendly and have predictable behaviors.

For example, if a robot is passing through a room with a human in it, it should look at that person so that they know the robot knows they are there.

"If it's just (seemingly) moving blindly around, I might become afraid of it, rather than realize that 'OK, it's perceiving that I exist and ... it's going to behave safely,'" he said.

The issue of safety is something Willow Garage's Cousins called vital to the future of home-use robots.

"Robots in human spaces ... have to be safe around people," he said, "and if you can't hit that bar, there's no way they're going to be acceptable."

In line with that, the PR2's 100-pound arms are held up by springs that allow them to "float," as opposed to powerful motors that would not only weigh a lot but also be capable of inflicting damage if the robot's programming were to malfunction.

"I'm a software guy and ... I don't trust software," he said. "So I want to make sure there's a backup plan in case the software goes crazy."

As robot-human interaction becomes more commonplace, questions about liability in the case of accidents have arisen in tandem.

Ryan Calo, an assistant professor at the University of Washington School of Law and an affiliate scholar at Stanford, has written about legal issues relating to robotics. He said current product liability law is adequate to cover single-use, "closed" robots that perform one task.

"They're supposed to do one thing, and they're supposed to do one thing safely ... (But) if you take a Roomba and you put a chainsaw on it, they can claim product misuse," he said. "You're not using it for the intended purpose."

But the legal issues get tricky with general-purpose robots that can be programmed by anybody, such as the PR2.

"People are going to do all kinds of wacky things with this technology, like they do with any technology," he said.

Calo said right now it's not clear how courts would rule if a general-purpose robot were to injure someone. But he said at the very least, the manufacturer would be a defendant and have to spend resources defending itself.

In a 2011 article for the Maryland Law Review, Calo wrote that open-robotics systems hold the most promise for innovation but that the legal uncertainty is a deterrent for companies to invest in the field.

"The law is basically stuck in this mode of product liability that's strict," he said. "So maybe we should immunize platforms for what users do with these so that (companies) have incentive to build the platforms because that will spur the robotics industry."

Building up the robotics industry was Willow Garage's primary goal, Cousin said, adding that the company expected to profit from this foundational work in the long term.

In line with this philosophy, PR2 customers who agree to share their programming code with the open-source community can receive a 30 percent discount off the robot's hefty price tag.

Cousins said the open-source nature of ROS jibed with Willow Garage's plan to build a robotics industry first and profit from it later.

In February, however, the need for profit apparently became a starker reality for Willow Garage, and Cousins announced that the company was changing its funding model and would "enter the world of commercial opportunities with an eye to becoming a self-sustaining company."

The statement said the company would continue to provide support for the PR2. A representative of Willow Garage declined to comment further on the company's decision.

Freelance writer Bryce Druzin can be emailed at


There are no comments yet for this post