Unlike the many soft computing applications where it suffices to achieve a "good approximation most of the time," a control system must be stable all of the time. As such, if one desires to learn a control law in real-time, a fusion of soft computing techniques to learn the appropriate control law with hard computing techniques to maintain the stability constraint and guarantee convergence is required. The objective of the paper is to describe an adaptive dynamic programming algorithm (ADPA) which fuses soft computing techniques to learn the optimal cost (or return) functional for a stabilizable nonlinear system with unknown dynamics and hard computing techniques to verify the stability and convergence of the algorithm. Specifically, the algorithm is initialized with a (stabilizing) cost functional and the system is run with the corresponding control law (defined by the Hamilton-Jacobi-Bellman equation), with the resultant state trajectories used to update the cost functional in a soft computing mode. Hard computing techniques are then used to show that this process is globally convergent with stepwise stability to the optimal cost functional/control law pair for an (unknown) input affine system with an input quadratic performance measure (modulo the appropriate technical conditions). Three specific implementations of the ADPA are developed for 1) the linear case, 2) for the nonlinear case using a locally quadratic approximation to the cost functional, and 3) the nonlinear case using a radial basis function approximation of the cost functional; illustrated by applications to flight control.

Adaptive dynamic programming

A recurrent neural network, called the Lagrangian network, is presented for the kinematic control of redundant robot manipulators. The optimal redundancy resolution is determined by the Lagrangian network through real-time solution to the inverse kinematics problem formulated as a quadratic optimization problem. While the signal for a desired velocity of the end-effector is fed into the inputs of the Lagrangian network, it generates the joint velocity vector of the manipulator in its outputs along with the associated Lagrange multipliers. The proposed Lagrangian network is shown to be capable of asymptotic tracking for the motion control of kinematically redundant manipulators.

A Lagrangian network for kinematic control of redundant robot manipulators

A decentralized adaptive control system (ACS) for a four motor-generator four-wheel drive hybrid electric vehicle (HEV) is designed and its ability to deal with unknown tire dynamics, changing road surfaces, and vehicle loading is evaluated. A system composed of four separate adaptive controllers is designed to control the vehicle's speed, steering, side slip, and energy management system. A nonlinear simulation model for the vehicle dynamics and its power train components is developed and used to evaluate the performance of the ACS, while the vehicle is simultaneously turning and accelerating or braking under varying loading and icing conditions.

Adaptive control of a hybrid electric vehicle

This paper describes the design of a decentralized adaptive control system (ACS) for a hybrid electric vehicle and evaluates is its ability in dealing with unknown tire dynamics, changing road surfaces and vehicle loading. The vehicle's speed, steering, side slip, and energy management system are controlled by four separate adaptive controllers. A nonlinear simulation model for vehicle dynamics and power train components is used for evaluating the performance of the ACS. The evaluation takes place as the vehicle is simultaneously turning and accelerating or braking under a variety of loading and icing conditions.

Adaptive control of a hybrid electic vehicle

It is shown that a Hopfield neural network (with linear transfer functions) augmented by an additional feedforward layer can be used to compute the Moore-Penrose generalized inverse of a matrix. The resultant augmented linear Hopfield network can be used to solve an arbitrary set of linear equations or, alternatively, to solve a constrained least squares optimization problem. Applications in signal processing and robotics are considered. In the former case the augmented linear Hopfield network is used to estimate the "structured noise" component of a signal and adjust the parameters of an appropriate filter on-line, whereas in the latter case it is used to implement an on-line solution to the inverse kinematics problem.

Linear Hopfield networks and constrained optimization

The neural network based robotic arm control concept covers three areas, decentralized adaptive joint control, an inverse kinematics, and path planning. Included are new results from the decentralized adaptive joint controller. This joint controller uses neural networks to adapt a proportional-integral-derivative (PID)/PVA controller. The results show that neural networks allow for fast, accurate control. The authors have tested the joint controller in a robotic testbed simulation software. The neural driven inverse kinematic system has produced accurate performance. The results show that the overall system outperforms conventional methods. >

A neurocontroller for robotic applications

The development of applications using artificial neural networks has been greatly enhanced by having a toolbox/library of neural network paradigms. Some of the applications that have been implemented with the toolbox include signal processing, sensor fusion, robotics, control, and fault diagnosis. At its lowest level, the toolbox consists of a software library written in C and Ada. This library is accessible directly, through command line routines, or through a graphical user interface on a high-performance workstation. This library significantly reduces the time necessary to develop neural network applications. Through the graphical interface, no programming is required to prototype an application using the paradigms. A network module can be created by simply selecting a network type and dragging it into a work area. The network structure and parameters may be tuned by manipulating graphical widgets such as sliders and dials. Module I/O can be sent between multiple network or nonnetwork modules by creating graphical pipes. >

A neural network toolbox for application simulation

A decentralized adaptive joint neurocontroller for an n-joint robot arm is described. The joint neurocontroller consists of a PVA controller (feedforward proportional-velocity-acceleration controller) and a PD controller (feedback position-derivative controller) whose terms are adapted by a recurrent functional link neural networks. It uses positional information only. Simulation studies of a two joint robot arm have shown that this controller is stable and robust. The joint neurocontroller performed as well or better than one current adaptive controller technique and one current nonadaptive controller technique. >

http://ieeexplore.ieee.org/iel2/632/5900/00226878.pdf

A decentralized adaptive joint neurocontroller

The authors explored whether neural networks can improve telerobotic performance. The neural network design is based upon an innovative three-tier distributed control architecture. The neurocontroller was tested on two simulated two-joint robot arms which had different dynamics. The tests compared the performance of five controller configurations. The findings indicate that a decentralized adaptive neurocontroller performed as well as or better than standard adaptive and nonadaptive controllers. This approach to autonomous control and path planning circumvents the tradeoff between speed and desired levels of accuracy, stability, and robustness by generating optimal trajectories without sacrificing computational speed or robustness. >

Underwater robotic operations using a decentralized adaptive neurocontroller

A neural network based fault diagnosis system is being developed for use in aerospace systems in which a family of neural nets replaces an online simulation process A new neural network implementation of one of the model-based algorithms was developed The authors summarize a series of computer experiments designed to benchmark the performance of this neural network based fault diagnosis algorithm in an environment where the good components of the systems are only known up to a tolerance band The results indicate that neural networks can generalize around the data on which they were trained, yielding better performance for unforeseen inputs than traditional algorithms in the case of the fault diagnosis problem described >

M. Lothers

Papers

A neurocontroller for robotic applications

A neural network toolbox for application simulation

A decentralized adaptive joint neurocontroller

Underwater robotic operations using a decentralized adaptive neurocontroller

A neural network based approach to fault diagnosis in aerospace systems