Compensating the Tracking-Error of a Mobile Robot by On-Line Tuning of a Neural Network

Abstract

In this paper the problem of compensating the tracking errors of a mobile robot is considered. The errors taken into account are mainly due to the time delays associated with the hierarchical structure of a commercial mobile robot. A black-box design of the compensator, based on feedforward Neural Networks is considered. Three different control architectures are presented. The first attempt, based on a non-goal-directed inverse-modelling-structure approach, turns out to fail. Then, two goal-directed control structures are considered: the first is a Neural Network-based adaptive controller, that follows the changes of the operating conditions; the latter is based on a more complex Neural Network, that receives the operating conditions as inputs, so obtaining a fastreacting Neural controller, which highly improves the tracking performances of a mobile robot.