BACS breathes perception into robots

We are sitting in a soccer stadium and discover our neighbor sitting in the 10th row. We recognize him with no difficulty at all, even though he is wearing sunglasses and a cap in his club colors. Recognition processes like this work because the brain, sensory organs and nerve pathways are able to pick up stimuli and process them, the sensory organs and brain cooperate to bring about "constructive perception", and finally "social perception" occurs when language, culture and custom combine to shape the information that has been obtained. The ability to classify things appears to be a fundamental characteristic of human intelligence, and one that gives robots a real "headache". In situations in which a robot has no access to knowledge of a pre-defined environment, and pre-programmed control is therefore not possible, the robot will tend to fail miserably in its task. But it is precisely autonomous robots capable of acting in response to a given situation that could be of great use to humans.

Dealing with uncertainty
This is the focus of BACS (Bayesian Approach to Cognitive Systems), an Integrated Project under the 6th Framework Programme of the European Commission which has been allocated EUR 7.5 million in funding. This is the first time that the ETH Zurich coordinates an EU project which brings together researchers and commercial companies working on artificial perception systems potentially capable of dealing with complex tasks in everyday settings. The basis of this research is Bayes' theorem. Thomas Bayes was an English mathematician and Presbyterian monk who lived in the 18th century. The theorem named after him describes alternatives for calculating the likelihood of events occurring using conditional probability. It is a model for rational judgment when only uncertain and incomplete information is available. Bayes' theorem is applicable to all questions relating to learning from experience. In the 50 months for which the BACS collaboration will run, the ten project partners will use the theorem to model neuronal functions and cognitive processes. The aim is to gain a better understanding of perception and action in living beings, to optimize existing learning algorithms, and to realize intelligent artificial systems.

Robot systems that like taking decisions
The scientific work being carried out under BACS makes robots with new capabilities a real prospect: robots capable of handling incomplete information, analyzing their environment, acquiring context-specific knowledge, interpreting the data and, together with humans, taking decisions. Specific implementations with market potential are already planned. "We are aiming to develop a system that can assist drivers of passenger cars and trucks by employing probabilistic control functions and driving strategies. This should make driving safer for both drivers and pedestrians," explains Roland Siegwart, Professor of Autonomous Systems and the BACS project coordinator. "Another area of interest is 3D modeling and surveillance of safety-critical applications such as monitoring structural changes in buildings or mines and safety-relevant infrastructure elements in power lines." European industry can use Bayes' alternative calculation models to good advantage; they have applications both in major companies in the automotive industry and mobile telephony, for example, and in small and medium-size companies active in niche markets such as healthcare, inspection, monitoring or even market forecasts



Contact:

Professor Roland Siegwart
Autonomous Systems Lab (ASL)
Institute of Robotics and Intelligent Systems (IRIS)
ETH Zurich
Tannenstrasse 3
CH-8092 Zurich
E-mail: rsiegwart@ethz.ch