An improved autoassociative network for controlling autonomous robots

Abstmck One promising approach to neural network controlled robotics is the use of autoassociative networks. These networks learn to move a "sensor and effector" vector through a plausible state-space. This approach is, however, hindered by the intrinsically inefficient nature of autoassociative networks. This paper outlines a novel approach that greatly increases the efficiency and resolution of associative networks, and has other implementation benefits as well.