|
Robots that learn
An anthropomorphic robot, capable of manoeuvring itself with ease in our environment and of interacting with it, and capable of understanding our commands and executing them. A tireless and trusty mechanical ‘slave,’ in fact. This age old dream of humanity, which has fed so many science fiction novels is perhaps no longer so distant. Professor Justus Piater, of the University of Liège, is developing learning protocols for the grasping of objects within the framework of the European programme, PACO-PLUS. ![]() An objective which is easy to formulate, but which is a lot less so to carry out! ‘We already use sophisticated robots in industry or, for example, in hospital corridors to transport objects,’ picks up Professor Piater. ‘But these robots are limited to carrying out tasks which are perfectly defined at the beginning. Clearing a table is formidably complex! Which objects to grab? How? Where to put them? It is necessary for us to develop a robot which has a genuine generic understanding of our lives and our living spaces. And that is clearly no longer possible using the way of thinking in the 1950s and 1960s, at the beginning of the development of artificial intelligences: a comprehensive pre-programming of the robot, which contents itself with carrying out a list of very precise tasks. In order to have an autonomous robot operate amongst human beings we can no longer be satisfied with such a perspective: there is too much variation in real life! Moreover it takes a child years to learn how to become integrated in it. Therefore, if we do not want to abandon this idea of an autonomous robot living amongst us, we will have to do things differently. Instead of equipping a robot with knowledge, we have to equip it with the ability to learn the world and interact with it. Exactly as a human child does.’
|
|
|||||||||||||||||||||
|
||