Robust Control of Cooperative Manipulators

In this chapter we deal with the problem of robust position control for cooperative manipulators rigidly connected to an undeformable load. Design paradigms to solve force/position control problems have been established in the literature to improve the performance of the cooperation. In [8] a control strategy for cooperative manipulators was proposed based on the independence of the position and force controls. The applied force between the manipulator end-effectors and the object is decomposed into motion force and squeeze force, which must be controlled. In [7], the hybrid position/force controller proposed in [8] is extended to underactuated cooperative manipulators. Cooperative manipulators, like any electromechanical system, are subject to parametric uncertainties and external disturbances. In [4], a semi-decentralized adaptive fuzzy controller with H1-performance is developed for fully-actuated cooperative manipulators. In that work, the dynamic model is derived using the order reduction procedure proposed in [5] for constraint manipulators; only simulation results are presented to validate the controller’s performance. In this chapter, two nonlinear H1 control techniques based on centralized control strategies are evaluated for underactuated cooperative manipulators: H1 control for linear parameter varying (LPV) systems [9] and H1 control based on game theory [2]. These controllers are applied considering the control strategy proposed in [8], where the squeeze force control is designed independently of the position control. In these cases, the H1 performance index considers only the position control problem. The adaptive fuzzy-based controller for fully-actuated cooperative manipulators proposed in [4] includes the position and squeeze force errors in the H1 performance index. Following a similar approach, in this chapter we present a neural network-based H1 control for fully-actuated and underactuated manipulators [6]. We assume that a nominal dynamic model is available for the neural network to approximate the model. As in [4], the H1 performance index includes the position