Figure 1. The Field Integrated Design and Operations (FIDO) rover is being used in ongoing NASA field tests to simulate driving conditions on Mars.
You may not have realized it, but for the past three years you could have learned what it feels like to drive a vehicle across the surface of Mars. Since 1997, NASA has made it possible to command simulated missions of a Martian rover via the Internet.
NASA set up the program to obtain insights into how scientists might use the Internet to guide vehicles roaming across the surface in future missions to the red planet. Those missions might last at least a year -- far too long to detain a group of planetary specialists who have research and teaching responsibilities at NASA's Jet Propulsion Laboratory (JPL), which oversees Martian exploration. So the space agency developed the Web Interface for Telescience (WITS), a piece of software that links a vehicle designed for Martian travel with Internet users. In its present form the software gives anyone with access to the Web the ability to guide a simulated rover "in just the same way that the scientists designate where they want the real rover to go," said Paul Backes of JPL's automation and control section.
WITS showed its stuff in May, when groups of high-school students from across the U.S. logged on to help control NASA's Field Integrated Design and Operations (FIDO) rover. This summer, JPL announced that NASA had chosen the FIDO rover to fly in the Mars 2003 mission. The students drove it remotely as it explored a remote portion of central Nevada, collecting black-and-white and color images, near-infrared spectra, and close-up measurements. "The students had a real role helping us to complete a very complicated rover field test," said Raymond Arvidson, a Washington Univ. (St. Louis, MO) geologist who directed the field tests.
FIDO's field testing represents just one example of a growing trend in robotics: controlling robots via the Web. That trend builds on more than half a century of practical experience controlling robots at a distance. Developed in the 1940s to handle radioactive materials, remotely controlled robots now explore space, perform salvage work in the ocean depths, and clean up spills of hazardous substances. Soon, surgeons will be able to use robots to perform operations at distant locations. These uses have one factor in common. "Almost all these robots are accessible only to trained experts," said Ken Goldberg, associate professor of industrial engineering and operations research in the Univ. of California/Berkeley and co-editor of Robots on the Web: Physical Interaction Through the Internet, due from MIT Press in 2001.
Goldberg said the new class of remote robots differs in that they are available to the general public via the Internet. "Curious adults and children are now able to actively participate in remote exploration," he said. Activities made possible by Internet-based telerobotics range from painting and running toy trains to puzzling out mazes and interacting with equipment at the site of the Titanic. The field promises more than game playing. "It has tremendous implications for education, allowing students to actively explore remote environments," Goldberg said. "It also has implications for research, where laboratories can share access to expensive resources such as coordinate measuring machines or scanning tunneling microscopes."
Goldberg and his research team, then at the Univ. of Southern California, devised the first robotic control system on the Internet in 1994. Their Mercury project consisted of a four-axis robot with a camera and air nozzle set up over a sandbox. Remote viewers could dig for objects buried in the sand from their Web browsers by positioning their mice and clicking on them.
A team at the Univ. of Western Australia headed by Ken Taylor quickly followed Goldberg's lead. It went online late in 1994 with the Australian arm. This six-axis robot manipulator and gripper allowed users to build elaborate structures with colored blocks. More recent telerobotic projects include remotely controlled domestic robots that can check that the doors are closed and machines switched off; a system that used minuscule robots driven by Swiss watch motors; and the Telegarden, another installation developed by Goldberg's team that allows Web users to instruct a robot to plant and water seeds in a real garden in Austria's Ars Electronica Center.
Web robotics relies on fairly standard Internet technology, although advanced varieties depend strongly on such components as Java, the new language of the Web. WITS, for example, contains a dedicated database, a dedicated server, and a series of software tools related to the mission at hand. However, all types of robotic activity carried out through the Web encounter two fundamental problems: communication delays, and lack of security.
"Communication delays have become a crucial question," Goldberg said. "Not only is there a time delay over the Internet, but the delays vary." In that way, Web robotics differs from planetary robotics, in which technicians know just how long their commands will take to reach robots exploring remote worlds. Nevertheless, developers of Internet robotics use similar approaches to mitigate the problem. "The general area is called supervisory control," Goldberg said. "You have some autonomy on the robot. So when you send a command like 'pick that up,' the robot knows enough to figure out the details of getting the object you want. You don't have to worry about fine-grain control."
Dealing with the incalculable time delays of the Internet demands a more sophisticated approach. "Techniques have been developed that adapt to the available time delay and work smoothly," Goldberg said. A team at MIT, for example, has devised an approach that changes the 'stiffness' of the Internet robotic system. "Imagine controlling a robot as driving a BMW," Goldberg said. "If the time delay increases, it will become like driving a Cadillac."
Security is an inevitable worry in any proactive application of the Internet. "The distributed nature of the Internet, designed to ensure reliability by avoiding centralized authority, simultaneously increases the potential for deception," Goldberg writes in the introductory chapter of The Robot in the Garden: Telerobotics and Telepistemology in the Age of the Internet.1 "The capacity for deception is inherent in the Internet and particularly vivid in the context of telerobotics."
Figure 2. Telerobot developed by Ken Goldberg's team. Users can manipulate the robot via the web to plant and water seeds in a garden in Austria's Ars Electronica Center. Photo by Robert Wedemeyer.
Goldberg faced that issue soon after the Telegarden started up. "Someone rigged it up to water the garden all weekend, which flooded the lab," he recalls. His diagnosis: "You have to make the interface dummy-proof, and to prevent concerted hackers from doing damage. You can never be sure of security, but you are always thinking about how someone can break in."
One fresh approach to Web-based telerobotics may ironically help to solve the problem by exposing robots to more users rather than fewer. The novel technology is collaborative control, which could permit hundreds of users at different locations to control a single robot simultaneously. "One person controlling many robots is an old idea," Goldberg said. "We think this is new. We're very excited about many people controlling one."
Goldberg and his colleagues at Berkeley outlined the potential uses and the technical challenges of collaborative control in a paper they presented in April at the IEEE International Conference on Robotics and Automation. They posit a group of users working together to control a telerobot. "Each user monitors a different sensor and submits control inputs appropriate to that sensor information," the group wrote. "All of these inputs must be combined to produce a single control input for the telerobot."
One way to do that is "control fusion" -- that is, averaging all the inputs. The Berkeley team has designed a client/server system based on Java software that permits up to 30 users to share the controls for a single robot. As a practical implementation, the team has designed a telerobotic form of Ouija, the board game from Parker Brothers, Inc. Each registered player has his or her own planchette, a plastic disk that slides over a board marked with letters and words such as Yes and No. "The group of users poses a question," Goldberg said. "As each user concentrates, the planchette slides across the board to indicate an answer. Although we do not address the claim the planchette is influenced by supernatural powers, in many cases it is influenced by the conscious and unconscious movements of all participants."
How can collaborative control improve the security of telerobotics? Through strength of numbers, says Goldberg. By averaging out the commands of its many users, the system drowns malicious commands sent by an individual bent on mischief in a sea of "good" commands from benevolent users.
1. The Robot in the Garden: Telerobotics and Telepistemology in the Age of the Internet. Edited by Ken Goldberg, MIT Press, June, 2000.
A former science editor of Newsweek, Peter Gwynne is a free-lance science writer based in Sandwich, MA.