SCIENCE

Machines with vision

Canadians are creating sophisticated robots

MARK NICHOLS July 12 1993
SCIENCE

Machines with vision

Canadians are creating sophisticated robots

MARK NICHOLS July 12 1993

Machines with vision

SCIENCE

Canadians are creating sophisticated robots

MARK NICHOLS

In a typical working day, Alan Mackworth, a professor in the University of British Columbia’s computer science department, thinks deeply about ways to give machines the ability to see. He also gets to play with toy electric cars. The game is played with two six-inch-long Porsches and the car that can nose a squash ball past the other is the winner. It may sound frivolous, but the underlying purpose of the game is utterly serious: it is part of a thrust among scientists to develop robots with senses and intelligence. In Mackworth’s experiments, an overhead television camera acts as a vision system for the cars, with a computer providing information on what the camera sees. Analysing the data, the cars shift from defensive to offensive modes of operation as they steer around the tabletop playing field. Similar experiments are under way in robotics laboratories across Canada, as scientists in a wide range of disciplines explore ways of endowing machines with human capabilities.

The explosion of robotics research in Canada has developed in the past 10 years to the point where more than $60 million is currently being spent on programs in university laboratories. The purpose of it all: to develop a distinctive high-technology sector that will help Canada to survive in the postindustrial global economy.

To a remarkable degree, that plan, backed by Canadian governments and private industry, appears to be working. A shining example is Spar Aerospace Ltd. At the headquarters of Spar’s advanced technology systems group just outside of Toronto, officials heaved a collective sigh of relief last week after the U.S. House of Representatives voted 220 to 196 to fund a scaled-down version of the space station Freedom. When it begins operating in space in 2001, Freedom will carry a 57-footlong remote manipulator arm and a smaller device called a dextrous manipulator. The two robotic units, built by Spar and a consortium of Canadian firms, will help the station to function in space. Says Karl Doetsch, director gen-

eral of the space station program at the Canadian Space Agency: “Canada has a real international lead in this area.”

While Canadian expertise in robotics and the closely related field of artificial intelligence has been building steadily through the past two decades, researchers got a badly needed boost in 1983 with the establishment of the Toronto-based Canadian Institute for Advanced Research, which channels funding from government and private industry to researchers across the country. The in-

stitute, in turn, spawned PRECARN Associates Inc., a not-for-profit corporation backed by major Canadian industrial firms that is dedicated to research into robotics and intelligent systems—jargon for advanced computer-based operations. Currently, PRECARN is channelling $40 million into research in the field over a five-year period that ends in 1996. As well, PRECARN manages one of the programs launched in 1990 under the federal Network of Centres of Excellence initiative and is distributing $23.8 million over a fouryear period for research into robotics and intelligent systems.

The infusion of cash into robotics and related studies has put Canadian researchers

on a level playing field with their foreign counterparts. In the United States, funding for robotics research began “drying up during the late 1980s,” says David Miller, a robotics expert with the Mitre Corporation, a federally-funded research centre in McLean, Va. “Now Canadian scientists seem to be getting the money. There is a lot of interesting work being done in Canada.”

The surge of activity in Canadian laboratories ranges from work on computer-controlled systems designed to cope with emergencies in nuclear power plants to research into ways of using robotics to help physically handicapped children. Much of the work centres on robot vision. Allan Jepson, a University of Toronto professor of computer science, is working with other scientists to develop a vision system that would enable a robot to enter and work in areas where there are high radiation levels or other hazardous conditions. The robot is being designed for the publicly owned power utility Ontario Hydro, which operates a network of nuclear reactors. The prototype is a squat, three-

wheeled machine with a video camera mounted on top to act as the robot’s eyes.

Because the robot is designed to operate in known environments, simple images of the objects inside, along with a map, can be stored in a computer that serves as the robot’s brain. The computer identifies easy-to-recognize visual elements—edges and comers, for example—and constmcts a crude image of what it sees. Then it tries to match that image with the ones stored in its memory. In a test during the spring, researchers asked the robot’s vision system to identify several simple objects. Said Jepson: “It worked great.”

Other scientists are striving to lay the groundwork for more sophisticated robot vi-

sion systems. One problem with technology like Jepson’s is that, inevitably, the robot has to pause while its computer compares images of the external world with the ones in its memory. At the University of British Columbia, James Little, an associate professor of computer science, is trying to perfect a vision system that would operate almost as rapidly as a human’s by processing 180 million pieces of visual information per second in a powerful DataCube parallel processing computer. Little is addressing another tricky problem in robot vision. Using a robot head mounted on a table, he is trying to formulate an algorithm (a type of data-processing procedure) that would allow the robot to turn its head to follow a moving object—and distinguish between its own motion and that of the object it is tracking.

In Montreal, McGill University scientists are working on a vision system that is modelled on the operation of the human eye. According to Martin Levine, a professor of electrical engineering and director of McGill’s Research Centre for Intelligent Machines, the human eye takes in a wealth of data from the area that it is focusing on, but filters out much of what it sees on either side. This reason for this, says Levine, is data compression: because the eye filters out much of the peripheral data, that vastly reduces the amount of data that the brain has to process. Now Levine’s team is trying to develop a set of computer eyes that will behave in the same way. As well, scientists have found that the human eye tends to focus on symmetrical objects, such as circles or human bodies. Says Levine: “We’re looking to see if we can use that as a way of

helping a robot to find and track objects.”

As well as vision, Canadian robot scientists are also tackling such areas as balance and touch. At the University of British Columbia, Dinesh Pai, an Indian-born computer scientist, is developing an unusual robot shaped like a tetrahedron, a solid form with four triangular surfaces. With a leg at each of the tetrahedron’s four points, the robot is designed to operate on uneven terrain; even if it took a tumble, the robot would always land with three feet on the ground. At Queen’s University in Kingston, Ont., experimental psychologist Susan Lederman is helping engineers in an attempt to design a robot with a sense of touch.

And at McGill, Ian Hunter, a New Zealandborn associate professor of biomedical engineering, is rapidly carving out a major reputation in the field of micro-robotics. To study the behavior of muscle cells, Hunter during the 1980s began developing an instrument with quartz arms so thin at the ends they cannot be seen by the naked eye. Using the instrument through a computer-controlled system, says Hunter, “we felt the contraction of a single cell for the first time.” More recently, with the help of a $1.3-million grant from Martinex Inc., a Montreal-based research and development company, Hunter put some of the same principles to work in developing a high-precision instrument for eye surgery that can make movements as small as one one-hundreth the diameter of a human hair.

Canada’s push to develop superior expertise in robotics and intelligent systems has begun to pay off economically. At Spar Aerospace, technology developed for the U.S. space program has been used in such earth-

bound applications as remote handling systems for servicing nuclear reactors. As well, university robotics programs have helped to spawn companies that put esoteric technology to practical use. According to Levine, McGill’s robotics program has played a part in the establishment of at least five local companies, including Mayan Automation Inc., which manufactures vision systems for industrial use.

Still, some scientists say that corporations are often slow to recognize the value of advanced robotics. Peter Lawrence, a professor of electrical engineering at the University of British Columbia, completed development last year of a device that allows operators to manipulate the working parts of heavy construction or logging equipment through a single hand control. The patented device, called a Co-ordinated Control System, would replace the clumsy controls currently used on excavators and give the operator a more direct “feel” for the equipment he is using. So far, says Lawrence, heavy equipment companies have not shown much interest in his device, perhaps because of the costs involved.

That could change. Experts say that there is usually a lag time of five years or more before new technology makes its way into the marketplace. Given that, it may not be long before Lawrence’s ingenious control, Alan Mackworth’s experiments with computer vision for toy cars and other Canadian developments begin finding applications in a world where robotics seems certain to play an everincreasing role.

MARK NICHOLS