Project aims to accelerate development of personal robotics
When technologists talk about a “platform,” they mean a combination of standardized hardware and software that lets developers focus on creating applications, without having to worry about building all the underlying infrastructure. The Internet is an example. It lets people share blogs and videos with friends without having to, say, invent their own global network.
For every application in computer science there are two choices: platforms and standards or incompatibility and fragmentation. So far, the robotics community has leaned toward fragmentation, and the result has been hindered progress. Kenneth Salisbury, a professor of Computer Science and of Surgery, thinks it’s about time the robotics community had a platform like the personal computer.
“The PC provided a platform where innovators could conceive of new ideas like spreadsheets and word processing and games,” Salisbury says. “It was not until then that there was a reasonably accessible and affordable platform and sharing of code like hardware drivers and input/output capabilities.
“We believe robotics is at the same threshold where we’ve got a lot of the basic capabilities. We can build a hand, we can plan a collision-free path across the room, we can control an arm to move to a precise position quickly. But to get all of those things in one place is kind of like the first PC.”
So Salisbury leads the Personal Robotics Program with CS Assistant Professor Andrew Ng in coordination with the department’s STAIR project, and with CS graduate student Eric Berger and mechanical engineering graduate student Keenan Wyrobek. After about 18 months of work, they have developed a first prototype. It is remotely controlled now (eventually it will become more autonomous) but it is an early milestone on the way to releasing a practical, affordable and complete robotics platform.
The team’s hope for the platform is to allow fellow researchers and eventually all kinds of developers with great application ideas, to focus on creating robotics applications more easily. They will be able to work faster if they don’t have to fret about building arms, or getting the robot to grasp objects safely.
When robots are as standardized, modular, and as open to development as PCs, they might become just as versatile, Salisbury says. Imagine going to the electronics store to bring home software such as “Leaf Raker 3.0” to add to a robot that already can wash the car and fetch the mail.
Salisbury said he and the team designed the prototype to function alongside people, helping them in their everyday life. The project has even linked up with the Stanford Center on Longevity, which studies the myriad implications of the globe’s aging population. Robots that can help the elderly and aid in health care could be important innovations as people live longer.
“We took the typical human environment and picked out objects such as a frying pan, and doorknobs, or a jug of milk… everyday things,” Salisbury said. “We used that to decide how strong it should be and what its range of motion should be.”
To truly enable applications such as household helper, grocery shelf stocker, or any other case where a robot would mix with people, a platform must be physically safe and yet still capable. Finding a mechanical way to strike that balance has been a key innovation in developing the platform.
At first glance, the prototype looks a bit like the robot from the 1960s TV series “Lost in Space,” except without as much of a head. Like that TV robot, who would wave its arms and yell, “Danger, Will Robinson,” the most exciting physical aspect of the Stanford robot is its two gray, fabric-clad, dangling arms.
Traditional robot arms are built so that the robot can move to a position and hold it. Robots with heavy arms or with the responsibility for lifting heavy loads are designed to be powerful and are therefore unsafe brutes. Robots with light arms and no mission to move heavy things are left uselessly weak in the name of safety.
What Wyrobek and Salisbury realized is that the two jobs of a robot arm don’t have to be accomplished by the same hardware. If a robot arm were designed with independent support, such that the motors that move it somewhere aren’t also responsible for holding it up, those motors wouldn’t have to be all that strong to move heavy things. Consider how little force one needs to muster to push around someone suspended by the water in a swimming pool.
So the researchers designed arms for the robot that decouple supporting the arms from moving the arms. The arms are essentially counterweighted with springs, much like a garage door that is easy to lift despite its actual heft. The result is a robot with arms that can each hold about 12 pounds and yet be quite gentle. The robot can tap a person on the shoulder one moment and go lug around the vacuum cleaner the next.
Robots that merely move can be useful, but only for limited tasks in controlled situations such as welding car parts in a highly automated factory. Around people, robots will need to have the intelligence to handle more free-wheeling situations. For example, imagine developing a robot home health aide whose job is to bring a person pills and water. Among other things, that robot will have to be able to find its way to the kitchen, find a glass, know how to turn on the water, fill the glass to an appropriate level, and then find the person where ever he or she may be in the house.
The specific logic for each of those tasks will be conceived by developers, but the platform will make that developer’s life much easier by providing a host of basic software services being developed and integrated by Berger and Ng. These services essentially add up to an operating system. They include interfaces to the various sensors and actuators, and basic algorithms for navigation and other essential capabilities.
In the long term, Salisbury hopes that some applications built on the platform will make it popular, bringing economies of scale to bear on its cost. He estimates that in mass use the $50,000 prototype would cost something closer to $10,000-$15,000.
Meanwhile, as developers made the robots more intelligent, the machines would gradually take more and more responsibility for their tasks. At first they’d largely be human-controlled but eventually people would only have to supervise them, he says.
“It’s going to be a while before we can say ‘robot, do the dishes,” he says. “But there may be an intermediate point where somebody can say, take that dish and put it on that shelf.”
And then sip a lemonade while “supervising” how it mows the lawn.