Helping robots cope with uncertainty

Brussels, 13 Jul 2006

Robots are getting smarter all the time, and are now able to perform highly complex activities, yet there are still large numbers of tasks which are second nature to humans but leave robots largely stumped. For example, we are able to recognise someone we know in a large crowd, even if they are wearing sunglasses and a hat. In contrast robots would simply be unable to carry out this task, as they are ill equipped to handle unpredictable situations where they do not have full knowledge.

The BACS (Bayesian Approach to Cognitive Systems) project, which is funded under the information society technologies (IST) section of the EU's Sixth Framework Programme (FP6), aims to design an artificial system which would enable robots to cope with a real world environment, where uncertainty and unfamiliarity are the order of the day. At the heart of the project is Bayes' theorem, which provides a model for making rational judgements when only uncertain and incomplete information is available. It lends itself particularly well to questions relating to learning from experience.

Humans are particularly good at responding to unpredictable situations and taking decisions without having all the facts. The project partners will exploit this fact by looking at humans and rats responding to realistic situations, and analysing the processes and neural pathways involved. They will then map these onto an artificial cognitive system to create robots which are able to handle incomplete information, analyse their environment, interpret the data and take decisions.

The resulting system could have a range of useful applications. Devices already exist which warn car drivers when they are too close to the car in front. A 'virtual co-driver' could extend this concept much further to improve road safety. While the human would remain in control of the car, the co-driver would monitor the driver's reactions to the traffic around them, the road, potential obstacles and other factors. If the driver starts to make mistakes in their driving, for example if they have fallen asleep at the wheel, the virtual co-driver could detect this and over-ride the actions of the human driver.

'This should make driving safer for both humans and pedestrians,' said Roland Siegwart, Professor of Autonomous Systems at the Swiss Federal Institute of Technology.

Another area where the researchers plan to learn from human abilities is surveillance. We have a natural ability to identify the unusual in a situation. By exploring how we do this, the researchers hope to create security robots which could identify an attack or unexpected presence.

The 10-partner project is coordinated by the Swiss Federal Institute of Technology in Zurich and is due to run until 2010.

Further information

CORDIS RTD-NEWS/© European Communities, 2006
Item source