NEWARK, DE. -- Flobot the Robot can already hover on a cushion of air, manipulating objects with magnetic grippers to simulate hands-off satellite repairs in space, and the machine may soon learn to perform such tasks optimally, up to 10 times faster, thanks to a new algorithm described in the December 1996 issue of the Journal of Dynamic Systems, Measurement, and Control.
Methodical, slow-moving robots -- in space, or on the ground -- may become relatively `smart' decision-makers if they're equipped with optimization algorithms that use "transformations," a mathematical device akin to a mental short-cut, says Sunil K. Agrawal, director of the Mechanical Systems Laboratory at the University of Delaware.
"When a robot tries to grip a satellite in space, it tends to rotate unpredictably because it is floating freely in a weightless environment," says Agrawal, one of 30 promising young researchers to receive a $500,000 Presidential Faculty Fellow Award from President Bill Clinton in 1994. "Predicting how a robot will behave in space -- and programming it to perform satellite repairs quickly -- is a real challenge."
That's why Agrawal's research team created Flobot, a wireless 170-pound robot capable of grabbing a moving object while hovering over a table. As jets of air are released from a tank on the robot, it moves over the table and starts searching for a targeted object. Using an overhead vision system, a master computer in Agrawal's laboratory tracks the robot and directs its movements via radio signals, explains graduate student Mahu Annapragada. Two long arms, each equipped with four motors, latch onto an uncooperative mock satellite when it moves within range of the robot's grasp.
In the past, Agrawal explains, researchers have tried to simulate the behavior of free-floating space robots by launching machines into orbit, or by dropping test capsules into a deep hole in the earth, to mimic weightless conditions. But space-based experiments are costly, he notes, and test capsules fall so rapidly inside tunnels that researchers may have a tough time generating accurate measurements. By comparison, Agrawal says, "Flobot is a low-cost alternative for practicing the tasks involved in repairing a satellite in space."
NEW MATH FOR SMARTER ROBOTS
Highly sophisticated robots are now capable of making high-speed maneuvers while traveling from Point A to Point B, Agrawal says. To achieve "real-time" control, he adds, a robot must constantly process and update mathematical inputs based on its current position, taking into account the speed of its movements. Unfortunately, he says, controlling and optimizing robotic movements using a classic mathematical tool known as "Lagrange multipliers" is an extremely computer-intensive task.
But Agrawal says his new technique of dynamic optimization using transformations could result in a 10-fold reduction in calculation time. This may help robots generate optimal trajectories in real-time, he says. "By using mathematical transformations instead of Lagrange multipliers," he adds, "I can explicitly embed a knowledge base, in the form of dynamic equations, into the robot's programming codes."
Annapragada is already using the new method to improve some of Flobot's movements. Agrawal hopes the programming technique can eventually be incorporated into all of Flobot's control systems.
BETTER CONTROL FOR EARTH-BOUND ROBOTS, TOO
Along with Flobot, Agrawal and his team of six graduate students are working on robots capable of performing dangerous or complicated tasks on earth. For example, a Spine Robot, designed to move like a human spine, could be perfected to serve as a probe, or perhaps to perform repair-and-retrieval tasks inside small holes in buildings and tunnels in the ground. Because Spine Robot's "vertebrae" segments are driven by many independent motors, Agrawal says, the machine is flexible enough to navigate narrow, curving spaces.
Snake Robot, a close cousin of Spine Robot, may prove ideal for zipping around obstacles on a factory floor, or for entering potentially hazardous environments. "When we were successful in developing Spine Robot," Agrawal says, "we added wheels to each of these individual vertebrae segments to create a Snake Robot. Compared to some existing robots now being used in factories, we think Snake Robot may be more adept at carrying objects back and forth through a cluttered industrial environment, or for dangerous inspection tasks -- say, inside a nuclear reactor."
A Hopping Robot is now in the preliminary design stage. If the robot can be programmed to respond to impulses from the ground, Agrawal says, "we might be able to invent an incredibly fast running machine."