Power and limits of reactive agents

This paper shows how reactive agents can solve complex tasks without requir ing any internal state and demonstrates that this is due to their ability t o coordinate perception and action. By acting (i.e., by modifying their pos ition with respect to the external environment and/or the external environm ent itself), agents partially determine the sensory patterns they receive f rom the environment. Agents can take advantage of this ability to: (1) sele ct sensory patterns that are not affected by the aliasing problem and avoid ing:those that are; (2) select sensory patterns in which groups of patterns requiring different answers do not strongly overlap; (3) exploit the fact that, given a certain behavior, sensory states might indirectly encode info rmation about useful features of the environment; (4) exploit emergent beha viors resulting from a sequence of sensory-motor loops and from the interac tion between the robot and the environment. The relation between pure react ive agents and pure representational agents is discussed and it is argued t hat a large variety of intermediate cases between these two extremes exists . In particular, attention is given to the case of agents that encode in th eir internal states what they did in the previous portion of their lifetime which, given a certain behavior, might indirectly encode information about the external environment. (C) 2002 Elsevier Science B.V. All rights reserv ed.