Neural network based expectation learning in perception control: learning and control with unreliable sensory system

摘要

In this article, we investigate the viability of our proposed neural network-based extension of the perception control concept introduced by Randl?v and Alstr?m. In their work, each of the expectation elements is linearly acquired such that the expectation gives only the dominant information of the recent past. This handicap could become a serious problem when the perception process is applied to real physical systems. Such an approach has no capability to sense the trend or the dynamics in the information. Here, we introduce an extension of the perception control process by using a radial basis function feedforward neural network to learn the trend and the dynamics in the information queue. Through our simulations, we show that our neural network-based method is better than the conventional method.