Method of Controlling and Correcting Variations in Parameters of Elements of Neural Network Decision-Making Components within Robotics Systems

This article proposes and investigates a method of optimizing the reliability of robotics systems designed via the use of the neural network components of information processing. The method suggests creating a fault-tolerant architecture of the neural network decision-making components as part of robotics systems. Inside this architecture there are some automated processes that control and correct any negative variations in the parameters of computing elements, caused by their partial or full failures due to external and internal destabilizing impacts. The target of this experimental research into the method was the computer model of a robotics system where a neural network decision-making component enabled three main functions to be performed: the computed path control, the response to any obstacles on the path and the detection of any stop points on route in order to take a beneficial action. The research has proved the method to be efficient to ensure the maximum fault-tolerance of neural network components of information processing in robotics systems of various applications.