Making robots cleverer

Do you wish you had a car that drove you to work, or a robot that vacuumed your home without bumping into the kids or the dog? In his doctoral thesis, Karl Granström has solved some of the problems a robot might come up against.

Cars that drive themselves while keeping track of other cars, and robots that can work among us humans are two applications that are getting closer to being realised. In his doctoral thesis, Granström has developed a number of different algorithms that help a robot or a computer in a car to follow moving targets and also understand more or less what they see - the dimensions of the object - and how close it can get.

“Research around extended targets is fairly new, and has come on strongly over the last five to ten years,” says Granström, who will soon receive his PhD in Automatic Control Engineering at the Department of Electrical Engineering.

Previous research into how robots can automatically follow a target dealt with punctiform targets. That works in certain circumstances, but not when robots begin to interact a little more in an everyday environment. For example, we humans have no trouble differentiating between a large, perhaps overweight adult and two much thinner people walking close together. This is much trickier for a robot.

Using a laser scanner, however, the robot can gather so many measurements that it gets a good understanding of how big the person or group of people is and approximately how close it can go.

It is often not so important for the robot to know how many people the tight group consists of. The problem arises when one person in a group suddenly splits off - a child breaks off and dashes out into the street right in front of an oncoming car.

If those systems allowing cars and robots to be autonomous are to work – in the sense that they are able to move without human supervision – they must also be able to quickly deal with this sort of problem.

In his doctoral thesis, Granström has developed methods and algorithms that make it possible, using technology called Probability Hypothesis Density (PHD) filters, to follow several separate targets, and that also interpret signals correctly if a child splits from a group of people or another car suddenly leaves the long stream of cars.

Karl worked principally together with fellow researcher Umut Orguner, who began as a postgraduate in the Division of Automatic Control Engineering, roughly at the same time as Karl began his doctoral studies.  The result was a series of articles that were also gathered together in the extensive thesis, entitled “Extended target tracking using PHD filters”.

Where he’ll be heading after his doctorate remains to be seen.

“I'll be staying at automatic control engineering for a couple of months, thinking about whether I want to do a postdoc, or start working in industry, a job a bit closer to development would be nice, but I haven’t really made up my mind,” he says.