Design On The Intelligent Control Of Rigid Manipulator With Position Constraints Based On Reference Pattern

Manipulator has been widely used in industrial production,space exploration,service industry and other fields due to its advantages of high production efficiency,strong durability and low production cost.With the increasing complexity and diversification of application scenarios and tasks,people put forward higher requirements for the intelligence of manipulator.The organic integration of pattern recognition and intelligent control is expected to improve the intelligent autonomous control ability of manipulator.Nowadays,the pattern-based intelligent control of manipulator has become a cutting-edge and challenging hot research topic.In this thesis,by using deterministic learning theory,the intelligent control of rigid manipulator with position constraints is developed based on reference pattern.In the online stage,for a class of rigid manipulator system with unknown dynamics and position constraints,the experienced knowledge and reference patterns under different tracking tasks are acquired and stored by learning the unknown closed-loop dynamics and modeling the reference dynamical models.In the offline stage,the manipulator can invoke the experienced knowledge,thus realizing the intelligent autonomous control with no human intervention.The main works of this thesis are concluded as follows:Firstly,for the rigid manipulator with unknown dynamics and position constraints,a neural network tracking control scheme is presented in this thesis.Through the state transition,the manipulator system with position constraints is transformed into an unconstrained equivalent system,in which an adaptive neural network controller is designed for the unknown dynamics.By using deterministic learning mechanism,the neural network will learn the experienced knowledge of the unknown closed-loop dynamics under this tracking task.It can save and reuse the knowledge in the form of constant weight matrix,so as to construct the learned controller under this tracking task.On this basis,the knowledge of the closed-loop dynamics of the manipulator under different tracking tasks is learned respectively.And then a set of learned controllers corresponding to different tracking tasks are constructed.Secondly,aiming at the identification and recognition of the reference trajectory,this thesis defines a reference trajectory as a dynamical pattern according to the dynamical pattern recognition method.In the online stage,a neural network identifier is constructed to identify the reference model,and the identification result can be stored in the form of a constant neural network as a reference pattern.By identifying different reference trajectories,a corresponding reference pattern library can be established.In the offline stage,considering that the current task is unknown,dynamical estimators are constructed by using the stored constant neural network,and the corresponding state synchronization estimation errors are obtained.Based on the above estimation errors,a recognition strategy combining threshold judgment and minimum residual error can be adopted to quickly recognize the pattern changes.According to the recognition strategy,a controller smooth switching strategy based on the obtained candidate learned controller group is proposed to reduce the chattering degree of the control signal in the switching process of reference patterns.Finally,a position constrained rigid manipulator intelligent control platform based on reference pattern is built by combining V-REP and MATLAB software.The effectiveness and practicability of the pattern-based control scheme designed in this thesis are verified by dynamical experiments under the open-source platform.This scheme has good tracking performance in the multi-task environment where the reference trajectory is changed and can realize intelligent autonomous control without intervention.