Pattern-based autonomous smooth switching control for constrained flexible joint manipulator

Abstract

The paper proposes a pattern-based control scheme for a class of flexible joint manipulators with unknown dynamics and position constraints. The scheme consists of two phases: (i) online controller adjustment and task pattern identification; (ii) offline pattern recognition and controller calling. In phase (i), the constraint position vectors are transformed into unconstrained variables by using a transformation function. Subsequently, a set of static neural learning controllers are constructed for different reference patterns by employing experience weights obtained from stable adaptive neural control. Moreover, a dynamic estimator is developed to identify different reference patterns. The identified patterns are stored by constant NNs, thereby constructing a trained pattern library by combining with their corresponding static neural learning controllers. In phase (ii), the dynamic residual system is designed to recognize a tested pattern by comparing the size of residual error between the tested pattern and trained patterns. To avoid misjudgment, a novel recognition strategy is proposed by pre-recognition and recognition phases. When the tested pattern is recognized, the relevant experience-based controller strategy is recalled by an autonomous smooth switching technology. The proposed pattern-based control scheme has some advantages including accurate pattern recognition ability, small controller chattering, and highly autonomous control. Simulation studies on a 2-link flexible joint manipulator are implemented to show these advantages of the proposed scheme.