Internet-based remote control by using Adaline neural networks

In this paper, we present a remote control scheme for Internet-based teleoperation. This control scheme relies on the real-time estimation of concurrent roundtrip delays in order to optimally assign tasks between the user and the robot. For this purpose, we employ an adaptive linear (Adaline) neural network for which most conventional learning algorithms are infeasible since the computation is usually too intensive to be practical. To get around this problem, we introduce a novel learning algorithm that is based on the maximum entropy principle. Compared to traditional learning algorithms, the computing cost of this algorithm is very low, which makes it possible for the proposed neural network to be implemented on-line in real-time.