- 225 - Institute of Mathematics, Warsaw Technical University, Warsaw, Poland and Institute of Electrotechnics, Politechnika Swietokrzyska, Kielce, Poland Some problems of the learning neuronlike system for the control of robots Ryszard Gawronski and Bohdan Macukow 1. Introduction On the bases of many general data concerning motor control system in higher animals and man, many authors propose different structures for voluntary control system. In any case it is evident that such a structure may be divided in three or four levels. The highest level is a decision level which generates some rather simple signals triggering a set of processes which in the lowest level controls the appropriate muscle contractions that is voluntary movement. It is evident that between the above mentioned levels there exist some systems (structures) which take into account the reai situation, the body position and appropriate accelerations. The most important level generates a set of signals which controls alfa and gamma loops for every controlled muscle. In spite of many investigations end models proposed we know very few on the intermediate levels of muscles control system. We try to inves- tigate a model of a system which shows the possibility of such a hierarch- ical control, where the intermediate stage is a neuron-like adaptive net. Besides it would be very interesting to have a learning system for an artificial hand of robot which can learn to perform some stereotypic movements.