Visitors to University College’s East Hall last Thursday were met with an unusual sight—Lego robots. And the machines, created by undergraduate students in UC’s Cognitive Science program, were on the move, following tracks, avoiding obstacles, and even playing goal.
The students, enrolled in COG342H1 Issues in Cognitive Science II, worked together in groups of four to five, using commercially available Lego Mindstorms kits to build the robots. Each kit comes with a central processing unit or ‘brain,’ a variety of sensors for colour, light, distance, and touch, as well as an operating system and a simple, block-based programming language for making everything work together.
“One of the things students are exploring is the idea that you can get a lot for a little,” says course instructor Professor Jim John. “That is, you set up your programming in just the right way so that a relatively simple set of instructions can generate a really complex set of behaviours, behaviours that look intelligent,” he says.
“This is similar to what we're seeing in the work on neural networks, which we’ve heard a lot about lately with the Vector Institute,” a newly announced AI hub coming to Toronto and led by machine learning pioneer Professor Emeritus Geoffrey Hinton, a Google Engineering Fellow and UC faculty member.
In addition to programming, “students also had the engineering task of building the robot so that it stays together and so that its body is fitted to what it’s supposed to be doing,” says John.
Student Blair Bouchard was part of a team that designed a robot to follow a track and avoid, or pick up, objects in its way. “It’s a line-following, obstacle-avoiding slash object-deciding robot, meaning if the obstacle is an object with a certain smallness, it becomes a target object for the robot to collect rather than avoid, ” he says.
Such self-deciding behaviour was also displayed by a goalie robot built by Shaila Sardar and her group. In response to a ball coming its way, the bot moves from side to side, flicking its arms to block the ‘goal.’
“What’s cool about the goalie robot is, not only does it get the ball, but it also ends up behaving the way a real goalie does,” says John. “If you’ve ever seen a shoot-out, about half the time, the ball goes one way and the goalie goes another. It’s human-like behavior that the robot is pulling off. And these robots are very simple, basic versions of what you’d see in industry.”
Ultimately, this responsive, human-like behaviour is what companies like Google are seeking to replicate in commercial technologies like self-driving cars.
“You want the cars to make the judgements we would make,” says John. “That plastic bag is okay to run over, that kid is not. So how are you going to program the robot to make it get that? That’s what these students have been exploring.”