ADAM: ADaptive Autonomous Machine zyxw Daan zyxwvu C. van Oosten Dutch Open University, P.O. Box 2960 6401 DL Heerlen, The Netherlands e-mail:voo@ ouhi08.ouh.nl Lucas F.J. Nijenhuis en Andr6 W.P. Bakkers Electrical Engineering Faculty, Control Laboratory University of Twente, P.O. Box 217 7500 AE Enschede, The Netherlands e-mail: bks @rt.el.utwente.nl Wiek A. Vervoort Informatics Department, Systems Programming and Architecture University of Twente, P.O. Box 217 7500 zyxwvut AE Enschede, The Netherlands e-mail: vervoort@cs.utwente.nl Abstract zyxwvutsr This paper describes a part of the development of an adaptive autonomous machine that is able to move in an unknown world, extract knowledge out of the perceived data, has the possibility to reason, and finally has the capability to exchange experiences and knowledge with other agents. The agent is not pre-programmed by its designer but was given simple rules of l$e i.e. what is 1. Introduction It is a great challenge to develop creatures that show intelligent behaviour and have a learning capacity. Since many years researchers are trying to give (artificial) life to beings that have autonomous behaviour [l-9, 11-14, 17- zyxwvut 211. But all trials so far have a major drawback. The developers put in these creations too much of their own knowledge in more or less static structures. As a consequence the creations do not come to artificial life in an appropriate way. That is to say their learning capacity is limited and they cannot survive in a strongly changing environment. 0-8186-7695-7/96 $5.00 zyxwvutsrqp 0 1996 IEEE Proceedings of EUROBOT ’96 good and what is bad. By evaluating its sensor inputs these rules of life were transfomzed into a rule based reactive system. Simulations of the system showed that the agent is able to learn by its own experience. By representing the leamed knowledge in an appropriate way, the acquired knowledge could be judged on its effectiveness and also this knowledge could be shared with other, less experienced, agents. To meet this goal the following demands have to be met by the system: The system may not incorporate human world knowledge, everything the system learns must be based on own experiences. The agent starts as a “tabula rasa”. Learning how to move is based on the state of mind of the agent. This state of mind resembles natural feelings of creatures like happiness, loneliness, sadness, etc. In order to store the information the system learns, a “dynamic data structure” is needed that enables the system, not only to store all kind of information, but new behaviours as well. zyxwv 143