UBC Reports
October 3, 1996

Robo prof

Alan Mackworth builds machines that see, reason and act

by Charles Ker
Staff writer

Picture The Dynamites have finished their first two-on-two scrimmage and, like any conscientious soccer coach, Alan Mackworth is reviewing game tapes.

"They're trying to be reasonably intelligent, but for the most part they look a bit stupid right now," says Mackworth. "They keep getting in each other's way."

The team is gearing up for its inaugural tournament in Japan next summer. Mackworth's goal is to sharpen his players' perceptive powers, their ability to co-operate, reason and take advantage of opportunities; a tall order for humans, even more so for robots.

The Dynamites are remote-controlled toy cars--modified, six-inch racing Porsches to be exact. Their soccer pitch, resembling an enlarged pool table with raised edges, dominates one wing of UBC's Laboratory for Computational Intelligence (LCI). Above the pitch hangs a colour video camera hooked up to a modest-looking piece of computer hardware. This, however, is no video game.

Computers attached to the overhead camera analyse what's happening on the pitch 60 times a second and convey this visual information to separate off-board computers for each car. Players are continually assessing speed, direction, where they are in relation to partners, competitors, the ball and the goal. And then, of course, there's strategy.

"Can I get to the ball before the other guy? Should I back off and play more defensively? They're thinking all the time," says Mackworth.

As founding director of the LCI, Mackworth has watched it develop into what many consider to be one of the best laboratories for integrated intelligent systems anywhere. Under Mackworth's guidance, the lab has grown from a three-professor operation in 1981 focused on computational vision into a team of eight professors building hybrid systems in mobile robotics, telerobotics, remote sensing, object recognition, decision making and computer reasoning.

Mackworth became interested in artificial intelligence while at Harvard in the late 1960s. He was pursuing a master's degree in applied math with the intention of going into mathematical psychology. Then he came across a collection of papers called Computers and Thought which laid out some philosophical arguments about whether only humans can think. Mackworth saw the readings as a personal challenge to him to build a thinking computer.

Says Mackworth: "Some philosophical questions you can just argue forever and never get anywhere. But a question like `Can computers think?' can be settled by building one. It was something I thought I could see in my lifetime and I wanted to contribute."

The professor built his first thinking machine for his PhD in Artificial Intelligence at the University of Sussex.

In the early 1970s, computers' sense of the outside world was limited to reading the pulses off a paper tape or punch card. By hooking a TV camera up to a computer, Mackworth wanted to know if a machine could perceive through sight. He was specifically interested in finding out what knowledge a computer needed about the world in order to differentiate between things in its view.

Mackworth based his thesis on the argument that people use generic stored knowledge to interpret images. Using this theory, he proceeded to develop basic algorithms--sequential sets of instructions that computers use to solve problems--which allowed his computer to identify blocks in a simple sketch drawing.

This initial exploration into the so-called "blocks world" of artificial intelligence became the basis for Mackworth's present-day research into computational constraint-based intelligence--the notion of using constraints as the basis for looking at the world, understanding pictures of it and arriving at solutions to problems along the way.

When Mackworth joined UBC's Dept. of Computer Science in 1974, one of his first projects was the development of a computer program which could label features on maps. By the early 1980s, the ability of remote sensing programs to read maps, identify clearcut areas and various stands of trees proved a godsend to photo-interpreters in the forest sector who were being bombarded with huge amounts of image data from satellites.

At the more local level, Mackworth points to the yearly campus nightmare of scheduling classes into rooms as a classic constraint-based problem which his algorithms have helped solve.

However, his true passion lies with developing dynamic, hybrid systems which combine elements of computer science and electrical and mechanical engineering.

In 1984, the Canadian Institute for Advanced Research (CIAR) chose artificial intelligence and robotics as the first area it would fund. Since then Mackworth has played a lead role in establishing both the CIAR's program in the field as well as the Institute for Robotics and Intelligent Systems (IRIS), a Network of Centres of Excellence program.

In the LCI, Mackworth co-ordinates a team of colleagues, staff and graduate students in an IRIS computational perception project called Dynamo, short for Dynamics and Mobile Robots. Joining the Dynamites in this initiative are: Spinoza, a robot with stereo vision which enables it to navigate unaided around objects and sense their distance; and the Platonic Beast, a robot which can move on different terrains without getting stuck.

But Mackworth says these gadgets are just platforms for testing constraint-based theories--theories which he believes hold the key to a safer future.

As microprocessors proliferate at a dizzying pace, he claims the controls under which many systems operate are becoming more ad hoc. He says his systems and those of his colleagues--coded to deal with specific constraints or problems --are the best way to ensure safety whether on an airport runway or in an elevator.

"An artifical intelligence program is a working theory, one that does something as opposed to just sitting there on a piece of paper," he says.

"The claim we make is that constraint-based systems give you powerful engineering tools for designing and building safe systems where you can guarantee their components because they are specifically built to engineering criteria."

The Dynamo collection also holds great practical promise.

Watching the Dynamites buzz around the pitch, Mackworth foresees a time when households will be cleaned by tiny robotic vacuum cleaners thinking and working together. Last summer, a Carnegie-Mellon University team successfully built a computerized car which drove itself across the country under its own vision--a feat that has direct links with UBC's Spinoza.

A billion dollars of research money is also being spent in California developing intelligent highways which feature platoons of computer-guided cars able to change lanes, exit and monitor erratic drivers nearby--all elements of the Dynamites' game plan.

As for the Japanese soccer tournament next year, Mackworth says the competition should be fast and furious, unlike the recent chess match pitting the computer Deep Blue against the Russian world champion.

Says Mackworth: "In casual chess you can take as long as you want to make the next move. In our game, if you think too long, you're going to lose."