Dongsheng Guo

Zhang Dynamics and Gradient Dynamics with Tracking-Control Application

Via solving time-varying linear equations, this paper shows the Zhang dynamics (ZD) method. Besides, the gradient dynamics (GD) method, which was originally designed for constant problems solving, is generalized for time-varying linear equations solving. Then, the ZD and GD methods are exploited together to solve the tracking-control problem of a nonlinear system as a new application. Simulation results on the nonlinear system further demonstrate the feasibility of the ZD and GD methods for tracking control of nonlinear systems.

Pruning-included weights and structure determination of 2-input neuronet using Chebyshev polynomials of Class 1

A new type of feed-forward 2-input neuronet using Chebyshev polynomials of Class 1 (2INCP1) is constructed and investigated in this paper. In addition, with the weights-direct-determination method exploited to obtain the optimal weights from hidden layer to output layer directly (i.e., just in one step), a new structure-automatic-determination method called weights-and-structure-determination (WASD) algorithm is proposed to determine the optimal number of hidden-layer neurons of the 2INCP1. Such a WASD algorithm includes a procedure of pruning the proposed neuronet (after the net grows up). Numerical results further substantiate the efficacy of the 2INCP1 equipped with the so-called WASD algorithm.

Manipulability-maximizing self-motion planning and control of redundant manipulators with experimental validation

For achieving optimal maneuverability, a performance index in the form of quadratic function is proposed and analyzed for the self-motion with manipulability maximization (SM3) of redundant manipulators. The corresponding SM3 scheme can automatically select the desirable configuration so that the manipulator is most flexible and best maneuverable. As joint-physical limits generally exist in an actual redundant manipulator, both joint-angle limits and joint-velocity limits are taken into consideration in the proposed SM3 scheme. For practical and protective purposes, a zero-initial-velocity constraint is also incorporated into the SM3 scheme to eliminate the large-initial-velocity weakness. The SM3 scheme can further be converted and unified into a quadratic program (QP). In addition, two very important bridge theorems are provided to guarantee that the QP can be solved by a numerical algorithm efficiently. By comparing with the scheme of the self-motion with middle-value approached (SM2VA), computer-simulation results based on five-link and seven-link robot manipulators demonstrate the effectiveness of the SM3 scheme. Furthermore, the experiment is conducted on an actual six degrees-of-freedom (six-DOF) push-rod-joint (PRJ) manipulator, which substantiates the effectiveness and physical realization of the proposed SM3 scheme.

More illustrative investigation on window-shaped obstacle avoidance of robot manipulators using a simplified LVI-based primal-dual neural network

One important issue in the motion planning of kinematic redundant manipulators is the online obstacle-avoidance. For such purposes, we proposed and unified the scheme formulation based on general quadratic-programs (QP), which incorporates physical constraints such as joint physical limits and collision-avoidance inequality. In this paper, a simplified primal-dual neural network based on linear variational inequalities (LVI) is presented for the real-time solution of such a collision-free inverse-kinematic planning scheme. The neural network solves the strictly-convex QP in an inverse-free manner, in addition to the simple piecewise-linear dynamics and global exponential convergence to optimal solutions. Further computer-simulations based on PA10 redundant robot manipulator substantiate the efficacy of the scheme formulation and its neural-network solver on window-shaped obstacle avoidance.

Growing-type weights and structure determination of 2-input Legendre orthogonal polynomial neuronet

In order to remedy the weaknesses of conventional back-propagation (BP) neuronets, a novel 2-input Legendre orthogonal polynomial neuronet (2ILOPN) based on the theory of the multivariate function approximation is constructed and investigated in this paper. In addition, based on the weights-direct-determination (WDD) method, two weights-and-structure-determination (WASD) algorithms with different growing speeds are built up to determine the optimal weights and structure of the proposed 2ILOPN. Numerical-study results further verify the efficacy of the proposed 2ILOPN equipped with the two aforementioned WASD algorithms.

Self-motion planning of functionally redundant PUMA560 manipulator via quadratic-program formulation and solution

In this paper, a scheme is proposed and investigated for the self-motion planning of PUMA560 robot arm, a functionally redundant manipulator (if the end-effector positioning is only considered). As joint physical constraints always exist in robot arms, we incorporate both the joint angle limits and joint velocity limits into the self-motion planning scheme as well. This scheme is formulated as a quadratic program (QP) subject to equality and bound constraints, and then solved online by using the LVI-based primal-dual neural network (LVI-PDNN). Computer simulations based on PUMA560 manipulator substantiate the efficacy of this self-motion planning scheme.

Zhang-gradient tracking controllers of Z1G0 and Z1G1 types for time-invariant linear systems

Zhang dynamics (ZD) and gradient dynamics (GD) are two types of powerful methods for online problems solving. In this paper, the tracking-control problem of time-invariant linear (TIL) systems is formulated and investigated. Then, by combining the ZD and GD methods, two novel types of tracking controllers are designed to solve the problem. Simulative results of TIL systems further demonstrate the feasibility and effectiveness of the ZD and GD methods for the tracking-control problem solving.

Weights and structure determination of feed-forward two-input neural network activated by chebyshev polynomials of class 2

Based on the theory of polynomial interpolation and approximation, a new feed-forward two-input neural network activated by a group of Chebyshev polynomials of Class 2 (i.e., TINN-CP2) is constructed and investigated in this paper. To overcome the weaknesses of conventional back-propagation (BP) neural networks, a weights-direct-determination (WDD) method is exploited to obtain the optimal linking weights of the proposed neural network directly. Furthermore, a new structure-automatic-determination (SAD) algorithm is developed to determine the optimal number of hidden-layer neurons of the TINN-CP2, and thus the weights-and-structure-determination (WASD) algorithm is built up. Numerical studies further substantiate the efficacy and superior abilities of the proposed TINN-CP2 in approximation, denoising and prediction, with the aid of the WASD algorithm which obtains the optimal number of hidden-layer neurons of the TINN-CP2.