We construct and examine a control system with a process called "perception
" that selects and completes information received from the system to be con
trolled. Through a number of basic control tasks, the suggested "perception
" paradigm is shown valuable when parts of the information from the system
being controlled sometimes are missing. Reinforcement learning is used to c
hoose the appropriate control action from the formed perception. (C) 2001 E
lsevier Science B.V. All rights reserved.