Multistage STM in a Multilayer Hebbian Learning Architecture for Local Navigation

In this paper we motivate and present a novel neural network architecture that includes multi-stage short-term memory (STM) and multilayer Hebbian learning. We apply this network as an adaptive steering assistant for an electrically driven wheelchair, which is equipped with tactile, sonar and other sensors. The influence of the adaptive controller increases with the probability that user commands result in collisions.