Deterministic Learning For A Class Of Rigid Robots With State Constraints

With the development of productivity and the popularization of automation technology,robot manipulators have been widely used in a variety of fields,such as industrial manufacturing,medical services,military and space exploration.Therefore,the research on the control problem of robot manipulators has aroused the widespread concern.Due to the highly non-linearity,strong coupling and uncertainty of robot systems,as well as the complex working conditions,it is extremely challenging for the control and learning of robot manipulators.As the main branch of intelligent control theory,neural network control has been widely used to solve the control problem of robot system with inherent non-linearity.Although neural network control has obtained very notable result in the research of manipulator systems,the learning problem of neural network control has not been solved well,which makes it difficult for manipulator to acquire,store and reuse the experience from stable closed-loop control system.Moreover,in practical applications,there are various constraints in manipulator control systems,such as task space constraints and velocity constraints.Violation of these constraints could result in serious degradation of system control performance,even damage the system and threaten the safety of the operators.Therefore,it is of great theoretical significance and practical value to solve the control and learning problem of robot manipulators with the predefined constraints.This paper firstly studies the neural network learning and control problem of a rigid robot manipulator system with the predefined tracking accuracy and velocity constraints.A performance function is used to characterize the transient and steady state performance of the tracking error,and a nonlinear transformation is used to convert the original manipulator systems into a nonlinear system without the constraints.Subsequently,the stored neural weights are used to design a static neural controller,which avoids the repetitive learning of unknown dynamics and improves the control performance of the manipulator system.Then,for the control problem of the rigid manipulator system with full state constraints,an adaptive neural network controller is proposed in this paper by adopting the barrier Lyapunov function(BLF)method,which ensures the closed-loop stability without violating the state constraints.Since the BLF scheme may cause the conservative of controller design and cannot achieve theaccurate approximation of unknown system dynamics,a new control scheme for rigid manipulator systems with full state constraints is proposed in this paper.In order to solve the above two problems,we firstly use a new transformed function to transform the constrained state directly.As a result,the bounds of the intermediate variables are not required to be known a priori and the conservative of the controller design is reduced correspondingly.Then an adaptive neural control scheme is proposed for the transformed nonlinear system and all the closed-loop signals are verified to be uniformly ultimate bounded.Finally,the neural learning controller is designed for similar control tasks by using deterministic learning theory,which achieve a better control performance.In this paper,a two-link rigid manipulator is taken as an example for the simulation and the simulation results show the effectiveness of the proposed scheme.