Yunong Zhang

Design and analysis of a general recurrent neural network model for time-varying matrix inversion

Following the idea of using first-order time derivatives, this paper presents a general recurrent neural network (RNN) model for online inversion of time-varying matrices. Different kinds of activation functions are investigated to guarantee the global exponential convergence of the neural model to the exact inverse of a given time-varying matrix. The robustness of the proposed neural model is also studied with respect to different activation functions and various implementation errors. Simulation results, including the application to kinematic control of redundant manipulators, substantiate the theoretical analysis and demonstrate the efficacy of the neural model on time-varying matrix inversion, especially when using a power-sigmoid activation function.

A unified quadratic-programming-based dynamical system approach to joint torque optimization of physically constrained redundant manipulators

In this paper, for joint torque optimization of redundant manipulators subject to physical constraints, we show that velocity-level and acceleration-level redundancy-resolution schemes both can be formulated as a quadratic programming (QP) problem subject to equality and inequality/bound constraints. To solve this QP problem online, a primal-dual dynamical system solver is further presented based on linear variational inequalities. Compared to previous researches, the presented QP-solver has simple piecewise-linear dynamics, does not entail real-time matrix inversion, and could also provide joint-acceleration information for manipulator torque control in the velocity-level redundancy-resolution schemes. The proposed QP-based dynamical system approach is simulated based on the PUMA560 robot arm with efficiency and effectiveness demonstrated.

From Zhang Neural Network to Newton Iteration for Matrix Inversion

Different from gradient-based neural networks, a special kind of recurrent neural network (RNN) has recently been proposed by Zhang for online matrix inversion. Such an RNN is designed based on a matrix-valued error function instead of a scalar-valued error function. In addition, it was depicted in an implicit dynamics instead of an explicit dynamics. In this paper, we develop and investigate a discrete-time model of Zhang neural network (termed as such and abbreviated as ZNN for presentation convenience), which is depicted by a system of difference equations. Comparing with Newton iteration for matrix inversion, we find that the discrete-time ZNN model incorporates Newton iteration as its special case. Noticing this relation, we perform numerical comparisons on different situations of using ZNN and Newton iteration for matrix inversion. Different kinds of activation functions and different step-size values are examined for superior convergence and better stability of ZNN. Numerical examples demonstrate the efficacy of both ZNN and Newton iteration for online matrix inversion.

A distributed multi-agent environment for product design and manufacturing planning

This paper describes a multi-agent approach to the integration of product design, manufacturability analysis, and process planning in a distributed manner. The objective is to develop a distributed concurrent engineering system to allow geographically dispersed entities to work cooperatively towards overall system goals. In the paper, an agent-based concurrent engineering system concerning product design and manufacturing planning, and its fundamental framework and functions are presented. The proposed model considers constraints and requirements from the different product development cycles in the early development phases and fully supports the concept of design-for-manufacturability. This methodology uses conflict resolution (CR) techniques and design-improvement suggestions to refine the initial product design. The model comprises a facilitator agent, a console agent and six service agents. Each service agent is used to model different product development phases, and the console agent acts as an interacting interface between designers and the system, while the facilitator is responsible for the decomposition and dispatch of tasks, and resolving conflicts of poor designs. A prototype system for part design, manufacturability analysis, and process planning has been implemented. The performance of the prototype system shows that it could be extended to include other service agents, such as assemblability analysis, to become a comprehensive distributed concurrent engineering system for geographically dispersed customers and suppliers.

Time-series Gaussian Process Regression Based on Toeplitz Computation of O(N 2 ) Operations and O(N)-level Storage

Gaussian process (GP) regression is a Bayesian nonparametric model showing good performance in various applications. However, its hyperparameter-estimating procedure may contain numerous matrix manipulations of O(N3) arithmetic operations, in addition to the O(N2)-level storage. Motivated by handling the real-world large dataset of 24000 wind-turbine data, we propose in this paper an efficient and economical Toeplitz-computation scheme for time-series Gaussian process regression. The scheme is of O(N2) operations and O(N)-level memory requirement. Numerical experiments substantiate the effectiveness and possibility of using this Toeplitz computation for very large datasets regression (such as, containing 10000~100000 data points).

Performance Analysis of Gradient Neural Network Exploited for Online Time-Varying Matrix Inversion

This technical note presents theoretical analysis and simulation results on the performance of a classic gradient neural network (GNN), which was designed originally for constant matrix inversion but is now exploited for time-varying matrix inversion. Compared to the constant matrix-inversion case, the gradient neural network inverting a time-varying matrix could only approximately approach its time-varying theoretical inverse, instead of converging exactly. In other words, the steady-state error between the GNN solution and the theoretical/exact inverse does not vanish to zero. In this technical note, the upper bound of such an error is estimated firstly. The global exponential convergence rate is then analyzed for such a Hopfield-type neural network when approaching the bound error. Computer-simulation results finally substantiate the performance analysis of this gradient neural network exploited to invert online time-varying matrices.

Repetitive motion of redundant robots planned by three kinds of recurrent neural networks and illustrated with a four-link planar manipulator's straight-line example

In this paper, a dual neural network, LVI (linear variational inequalities)-based primal-dual neural network and simplified LVI-based primal-dual neural network are presented for online repetitive motion planning (RMP) of redundant robot manipulators (with a four-link planar manipulator as an example). To do this, a drift-free criterion is exploited in the form of a quadratic performance index. In addition, the repetitive-motion-planning scheme could incorporate the joint physical limits such as joint limits and joint velocity limits simultaneously. Such a scheme is finally reformulated as a quadratic program (QP). As QP real-time solvers, the aforementioned three kinds of neural networks all have piecewise-linear dynamics and could globally exponentially converge to the optimal solution of strictly-convex quadratic-programs. Furthermore, the neural-network based RMP scheme is simulated based on a four-link planar robot manipulator. Computer-simulation results substantiate the theoretical analysis and also show the effective remedy of the joint angle drift problem of robot manipulators.

Zhang neural network solving for time-varying full-rank matrix Moore–Penrose inverse

Zhang neural networks (ZNN), a special kind of recurrent neural networks (RNN) with implicit dynamics, have recently been introduced to generalize to the solution of online time-varying problems. In comparison with conventional gradient-based neural networks, such RNN models are elegantly designed by defining matrix-valued indefinite error functions. In this paper, we generalize, investigate and analyze ZNN models for online time-varying full-rank matrix Moore–Penrose inversion. The computer-simulation results and application to inverse kinematic control of redundant robot arms demonstrate the feasibility and effectiveness of ZNN models for online time-varying full-rank matrix Moore–Penrose inversion.

A set of nonlinear equations and inequalities arising in robotics and its online solution via a primal neural network

Abstract In this paper, for handling general minimum-effort inverse-kinematic problems, the nonuniqueness condition is investigated. A set of nonlinear equations and inequality is presented for online nonuniqueness-checking. The concept and utility of primal neural networks (NNs) are introduced in this context of dynamical inequalities and constraints. The proposed primal NN can handle well such a nonlinear online-checking problem in the form of a set of nonlinear equations and inequality. Numerical examples demonstrate the effectiveness and advantages of the primal NN approach.

Zhang Neural Network Versus Gradient Neural Network for Solving Time-Varying Linear Inequalities

By following Zhang design method, a new type of recurrent neural network [i.e., Zhang neural network (ZNN)] is presented, investigated, and analyzed for online solution of time-varying linear inequalities. Theoretical analysis is given on convergence properties of the proposed ZNN model. For comparative purposes, the conventional gradient neural network is developed and exploited for solving online time-varying linear inequalities as well. Computer simulation results further verify and demonstrate the efficacy, novelty, and superiority of such a ZNN model and its method for solving time-varying linear inequalities.