Live: 300+ Best Black Friday Deals Live: Black Friday TV Deals BF Deals Under $25 BF Deals Under $50 5 BF Splurges 8 BF Must-Haves 15 Weird Amazon BF Deals BF Cheat Sheet
Want CNET to notify you of price drops and the latest stories?
No, thank you

Baby steps for Dexter the robot

Start-up Anybots sees a day when robots will do many things that humans now do. For now, though, they're just working on staying upright. Photos: Meet Dexter and Monty, the Anybots Video: Humanoid robot trains to do the dirty work

It clanked. It whooshed. But most important, it walked, without falling down.

"It" is Dexter, an upright, humanoid robot taking shape and getting exercise at Mountain View, Calif.-based start-up Anybots, which wants to make robots more human. At the moment, that apparently means skeletal and a bit shaky on its feet, but baby steps are always the beginning.

Dexter is notable in several ways. First, it's designed as a humanoid at a time when most robots, at least in the U.S. and Europe, are anything but. Just consider machines like from iRobot, or the that emerged recently from the labs of a Swiss university.

Second, Dexter makes extensive use of pneumatic technology--that is, Anybots uses air cylinders to drive the arms and legs. That goes against the grain of current trends in the robotics industry.

"The main objection in robotics textbooks is that, compared to electric motors, pneumatics are hard to control precisely," said Trevor Blackwell, who started the Anybots project six years ago. "This is true. Like muscles, pneumatics are soft and imprecise. But with a big brain controlling them, muscles work pretty well, and I found that with the right software I could also get good results with pneumatics."

Third, and most ambitiously, Anybots wants its creations to be all-purpose, not specialized à la Roomba. The goal is nothing less than to create a robot that can be taught to do all the things humans can do. To date, --which, among other things, can push a wheeled serving tray and jog a little--have been very much works in progress.

"Most of the robots on the market today are fairly special-purpose, and they do one job very well. A humanoid with legs probably won't do any particular job better, but it can go more places and do more jobs than any particular robot," Blackwell said.

The biggest push for humanoid robots is in Asia, and especially Japan. The Japanese government has even set deadlines for helper robots in human environments. By next year, robots are expected to be able to straighten up rooms, by 2013 to make beds, and by 2016 to lift and carry the infirm. The , meanwhile, has learned to pour tea.

For now, there's still much work to be done. Dexter is remotely controlled by a human operator and has learning software that allows small adjustments and refinements. Anybots has been showing off Dexter and fellow robot Monty in a limited fashion (including a video on the company's Web site), but Blackwell thinks it'll take another two to three years to get past the prototype stage.

Start-up thinks robots may be used to take on tough tasks.

Others are also in the early stages of crafting humanoid helpers. At the Massachusetts Institute of Technology, for instance, researchers are developing a humanoid called Domo that can grasp objects and place them on shelves or counters.

"The real potential of robots in the future is going to be realized when they can do many types of manual tasks," Aaron Edsinger, an MIT researcher who has been working on Domo for the last three years, said in a statement this week.

As the tinkering progresses, so too does work on the business side of things.

The Anybots project is being funded by capital that Blackwell got from being a principal in start-up Viaweb, which created a point-and-click Internet storefront system and then in 1998 was acquired by Yahoo. He took on two employees along the way, and the company is now seeking external venture capital.

"We need to scale up and have people working on a bunch of different parts," said Blackwell.

Eventually, Anybots wants to sell its software and partner with manufacturers that can produce different parts of the robot, to lower overall costs.

Right now, the price tag can be daunting. The Anybots team expects that the robot can stay at $100,000. "It's really no more complex than a car, so it shouldn't cost more than a car," said mechanical engineer Scott Wiley, who joined the company three and a half years ago.

From experience the group knows that VCs tend to be enthusiastic but cautious, especially toward the robotics industry. "It's been overhyped and has gone through a few circles of hype. Therefore people are taking everything with a grain of salt," said Daniel Miller, vice president of engineering, who started working with Anybots only a few months ago. "But in 2001, did you ever think another Internet company would be sold for $3 billion? These things go in cycles. Investors will be people that are willing to take a bet."

Estimates vary as to the overall size of the market. The Robotics Industries Association says the overall robotics industry in North America is worth about $1 billion, and hasn't increased in some time--in fact, it declined last year. Worldwide, industrial robots--like those used in the automotive industry, for example--are valued at $5 billion, according to Dan Kara, president and founder of Robotics Trends. Other segments, like robotic toys and military robots, are valued at between $1 billion and $6 billion, depending on what is included in the figures.

"The data is not robust," Kara said.

Anybots aren't designed to be witty sidekicks, like the robot in Lost in Space. Instead, Blackwell envisions that they will perform dirty or dangerous jobs that are unattractive or too risky for humans, like working in environments with hazardous chemicals, moving boxes of papers into storage and helping in situations where one person has to be in several places at the same time. Essentially, market demands will help shape the company's labors.

"We want people to tell us what the robot should be used for," said Blackwell.

Still, the company is focusing on building robots that can work in places originally designed for people.

"People don't want to redesign their homes or offices around robots," Blackwell said. "Humanlike robots are made to fit the environment, rather than (people) having to change the environment to fit the robot."

Futurists, screenwriters and scientists for years have dreamed about wisecracking robots that can perform household tasks, but so far such machines haven't taken the world by storm. Legs, arms and artificial intelligence, it turns out, are tough to replicate. A first generation of functional robots has emerged in recent years, but they get around on wheels (as does Roomba or ) or tracks (for instance, iRobot's Packbot). Robots shaped more like animals and people have largely been relegated to , the most famous being Sony's Aibo.

James Kuffner, a professor at the Robotics Institute at the School of Computer Science of Carnegie Mellon University in Pittsburgh, is interested in humanoid robots' motion planning capabilities. He sees a future for humanoids both on wheels and then later on legs. They'll likely be a reality, he says, in 20 to 30 years.

"A lot of people ask: why would you want anthropomorphic robots? The practical answer is that a human-shaped robot is ideally suited for human environments," said Kuffner. He points out, for instance, their potential ability to use door handles, chairs, cockpits, machines and hand tools.

But that's no easy task. Kara of Robotics Trends says legs add a great deal of complexity to the robot, and it takes a lot of energy and advanced software to make a legged robot work. But as challenging as it is to build a walking robot, legs would have certain advantages in human environments--they could climb stairs, clamber over things or drive cars. It's all part of the dream of a general-purpose robot that can learn to do anything and could accomplish any task a human being can do.

"Having legs is part of that. It would be able to navigate in spaces that a wheeled robot would have trouble with," said Kuffner.

Anybots' Dexter started walking in February. One crucial improvement was to reduce the weight on its lower legs. Dexter weighs between 150 and 160 pounds with its arms attached, and 135 pounds without them. When Dexter falls, it tries to get back on track again.

"Most human-size robots powered by electric motors would never have survived the physical injuries Dexter has," said Blackwell. He says pneumatics handle impacts and falls much better than electric drives, which matters because in order to learn to walk, robots--like children--have to fall hundreds of times. In six months Blackwell thinks Dexter will run, and then eventually walk over more difficult terrain.

Eventually, Dexter will be merged with Monty, which has arms and stands on a wheeled Segway base. One arm has a gripper that can lift up to 35 pounds. The other arm has a hand with soft rubber touch pads on each finger that it can use to grab things. Monty also has 12 cameras in its head that monitor the action around it.

"We'd like to see a robot that can do a few things and become more like a personal computer. We think that people can teach it to do different things in different areas," said Miller. The robot should also be controllable over the Internet, the company says.

Kara paid Anybots a visit recently and thought it was enlightening to see a robot company focusing on software. "They seem to be moving progressively, and I have never seen a robot jump before."

Still, for all the positive signs, many uncertainties remain.

"It's difficult to say where Anybots is heading. They have got to get a number of prototypes working," said Kara.