The modeling of two cooperating underwater vehicles with on-board manipulators carrying a common rigid load is studied. The velocity relations and the kinematic coupling are discussed. The kinematic constraints are due to fact that the velocities of the manipulator end-effectors are related via the rigid load. For the entire system, a dynamical model that takes into account the hydrodynamic effects is established for the design of a controller. The dynamical model is based on the Lagrange formulation and is developed in the world coordinate frame. The dynamical coupling arises as a result of the forces and moments transmitted through the common load. Several properties of the system under consideration are also discussed.

Redundant manipulators have more degrees of freedom (DOF) than the DOF of the task space. This implies that the number of joint position variables is greater than the number of variables specifying the task. The problem of solving the kinematic equations for the joint variables is underspecified unless additional equations/constraints are introduced to obtain a well-posed problem. A dynamic level redundancy resolution is proposed. The joint space model is transformed to a reduced-order model in the pseudovelocity space. The elements of the foregoing transformation matrix indirectly determine the contribution of each joint to the total motion. These elements are selected using two fuzzy logic-based methods so as to minimize the instantaneous manipulator power: (1) in the velocity method, a nullspace vector in the velocity relationship between the two spaces is determined by imposing a constraint on the continuity of the joint velocities at the time instant when the elements of the transformation matrix experience a discontinuity and (2) in the torque method, an alternative approach introduced to reduce the computational complexity, the changes in the transformation matrix are made continuous with respect to time by the appropriate choice of a nullspace vector in the joint torque expression. Simulations are given.

The problem of a robotic manipulator catching a moving object (target) is viewed as pursuer-evader problem. It is formulated as an optimization problem to determine the max-min=min-max (a saddle point) solution. After the feedback linearization, the models in the world coordinate system for the manipulator and the object are assumed to be linear, and the performance criterion is quadratic. The continuous-time models are combined to a single vector differential equation constraint. The optimum control theory determines the optimal inputs to the manipulator and the object in the feedback form. The gains of the feedback control laws contain the solution to the Riccati equation. It is shown that the positive definiteness of this solution yields a requirement for the capture: The difference of the reduced controllability matrix of the manipulator and that of the object (target) must be positive definite.

Redundant manipulators have more degrees of freedom (DOF) than the DOF of the task space. The redundancy resolution can be based on the kinematic or the dynamic equations of the manipulator. In the article, a dynamic level redundancy resolution is proposed. The manipulator dynamical model in the joint space is first transformed to a reduced-order model in the pseudovelocity space. The elements of the foregoing transformation matrix indirectly determine the contribution of each joint to the total motion. These elements are selected using a fuzzy logic based method so as to minimize the instantaneous manipulator power. A nullspace vector in the velocity relationship between the two spaces is determined by imposing a constraint on the continuity of the joint velocities at the time instant when the elements of the transformation matrix experience a discontinuity. The application of the proposed method to resolve the redundancy is illustrated by simulations.

The transformations between the (Cartesian) p-space, the (joint) q-space of a manipulator and a (reduced dimensional) /spl mu/-space are first presented. On the basis of the reduced order model in the /spl mu/-space, a low-dimensional torque control is determined. For implementation, it is transformed into the q-space while constraints are taken into account. It is then demonstrated that the foregoing transformation can be determined so that a well-defined performance criterion is optimized. The proposed approaches are illustrated by simulations.

This paper addresses the control of a manipulator with link flexibilities. The increased complexity in its dynamics presents challenges to controllers based on non-colocated sensing. In this paper a nonlinear predictive control approach is presented using a discrete time multilayer perceptron network model for the plant. The neural network model is trained to predict future outputs based on the available past measurements. At each sampling instant, the discrete time control input is calculated by minimizing a performance criterion. The method is compared to non-model based collocated PD control. Simulation results are presented.

An automatic control system for backhoe type excavators during free motion and digging operations is presented. Some of the uncertainties associated with the basically unstructured environment of soil digging tasks are dealt with by using an adaptive control system capable of on-line learning and control of the dynamic response over a wide range of parameter variations. The proposed control system comprises a primary and a secondary controller; the former is used to linearize the plant and the latter to compensate for modeling errors. The primary controller is implemented as a feedforward multilayer neural net trained to emulate the inverse dynamics of the plant. The secondary controller is a PID controller with the gains tuned so as to provide a satisfactory transient behavior. Simulation results are used to demonstrate the applicability of the proposed control scheme.

Necessary conditions for the optimality can provide additional equations to determine the configuration of a redundant arm for a given pose of an end-effector. The optimal control for the motion of redundant manipulators that minimizes an accumulated performance criterion subject to the motion constraints is presented. Necessary conditions for optimality are given. Algorithms for the calculation of a numerical solution to the optimization problem are described. An example for a three-link planar manipulator is presented to illustrate the presentation.>

The visual information obtained from a camera and an image processing unit are incorporated in an adaptive control algorithm to make a robotic manipulator grasp a moving object. Because of the inherent time delay caused by the image processing, the motion of the moving target is predicted in real time and is used in the online planning of the trajectory for the manipulator motion. Since the dynamics of the target are assumed to be unknown, the prediction is accomplished by means of an autoregressive discrete-time model. On the basis of the predicted motion of the object, the planner determines online at each control sampling instant the desired trajectory point (subgoal) for the controller. The subgoal point is tracked by controlling the end effector with self-tuner until grasping occurs. A simulation study and laboratory experiments are presented to demonstrate the performance of this visual feedback system.>

A hierarchical classification procedure is proposed to classify nine classes of red oak boards, eight classes of surface defects, and a class of clear wood. It utilizes prior knowledge about surface defects and texture information. After the preprocessing, the dark and bright regions on the board images are separated. The background and the wane defect are identified using their locations as prior knowledge. The rest of the surface defects and clear wood are divided into three subsets: circle, line, and texture. The surface defects belonging to each subset are characterized using the mean of the gray levels and the CAR model parameters. The approach improves the resolution of the defect locations. It also reduces the computational time as compared to approaches based only on the texture information. The feasibility of applying this procedure to the images of wood boards in a dusty environment was investigated. It was found that the performance of the proposed hierarchical procedure deteriorates when the sample boards are at least partly covered by dust.>

The control of the motion of a two-link manipulator with a flexible arm is studied. The first link is rigid, and the second link has a flexible part as an extension of a rigid part. The motion of the manipulator takes place on a horizontal plane. The dynamics of the manipulator are determined in Lagrange's formulation. The positions (output) of specified points on the flexible link are obtained in the Cartesian base coordinate system using strain-gage and joint-variable measurements. The inputs are applied to the actuators of the revolute and prismatic joints. For the controller design, a time-series multivariate model of the autoregressive exogeneous (ARX) type is used to describe the input-output relation. The discounted least-square method is used to estimate parameters of the time-series model. A self-tuning controller (STC) is designed so that the positions of specified points on the flexible link track the given trajectory points. The controller operates on the Cartesian coordinates, specifying the positions of the chosen discrete points on the flexible link. Simulation results as well as laboratory experiments on a Stanford/JPL arm controlled by an STC are presented to illustrate the approach.>

A dynamical model and a control architecture are developed for the closed-chain motion of two N-joint manipulators holding a rigid object in a three-dimensional workspace. Dynamic and kinematic constraints are determined and combined with the equations of motion of the manipulators to obtain a dynamical model of the entire system in the joint space. Reduced-order equations of motion and a functional relation for the generalized contact forces are developed. The problem of solving the reduced-order model for the forward and inverse dynamics is discussed. Control laws are determined so as to decouple the force-controlled and position-controlled degrees of freedom during motion of the system.>

The task of raising a rectangular parallelopiped object on the plane surface to a standing position by a robot manipulator is considered. An initial contact between the object and the surface is assumed to be a point contact. To perform the task, the manipulator needs to be servoed to that the object position changes from a point contact to an edge (line) contact and then to a plane contact, using sensory feedback from tactile and force/torques sensors. After an initial contact, the object is rotated around the gripper approach vector till the object makes a line contact with the surface. The angle between the gripper-held object and the surface is then calculated from tactile sensory information. The position and orientation of the gripper are modified based on the current gripper position and orientation, and the angle calculated so as to place the object on the plane. A simple experiment is conducted to demonstrate the applicability of calculating the angle between the object and the surface from tactile sensory information.>

The classification of surface defects on wood boards using random field models is presented. The gray levels of images of wood boards are assumed to be governed by causal autoregressive models. The parameters of the model equations are used to form a feature vector that characterizes the board images. These parameters are estimated by two robust algorithms: an algorithm that combines the data cleaning procedure and the two-dimensional M-estimation method, and an algorithm based on the two-dimensional generalized M-estimation method. The estimated parameters are used in the classification of sample boards into the nine classes: eight types of surface defect, and clear wood (no defects). The construction of the hierarchical tree classifiers is described. The experimental testing of the constructed classifiers is described. The accuracy of the classification is discussed.>

An online algorithm is proposed to test interferences and generate collision-free trajectories for the motions of loosely coordinated multiple robot manipulators controlled independently by means of PD (proportional differential) controllers. A hierarchical multiple-robot-control architecture is proposed. The approach is demonstrated by simulation studied.>

The pseudolink concept for representing the tip position of a nonrigid link is developed. The pseudolink is defined as the line segment connecting the proximal and distal ends of the links. Pseudolink length and angle are determined online using joint encoder information together with strain measurements taken along the link. External sensors are not required and trajectory planning can be carried out a priori under a rigid-link assumption. A self-tuning control based on pseudolinks is discussed. Laboratory experiments on a rotating nonrigid link show the proposed control leads to improved performance over a control that ignores compliance.>

The force-position controller is designed for robotic manipulator in the Cartesian coordinate system. A time series model is used for the controller design. An adaptive controller with self-tuning is determined (i) by minimizing a well defined quadratic performance criterion (ii) by pole placement. Comparison is made between the performance of the two controller. Successful simulations of force-position control using both controllers with self-tuning are presented.

Some of the edge detection methods well known in robotic vision systems have been applied to detect edges in tactile images. Two dimensional median filtering with a 3 × 3 window size has been employed for the removal of noise present in the tactile images with excellent results. The LTS-200 tactile sensor system developed by the LORD Corporation has been used to obtain the tactile images analyzed in this research. The study has been carried out to obtain the contours of objects smaller in size than the LTS-200 sensor (with an active area of 1.131 in. × 0.707 in.) with a single probing operation. For larger objects, a straight line extraction method has been employed by which one can obtain the polygonal approximation of the shape of the object by sequential touching and integrating the straight lines so extracted.

The force-path control of a robotic manipulator is presented here in the joint and in the Cartesian coordinate systems. An autoregressive model with external excitation (ARX-model) is introduced for designing an adaptive controller with self-tuning. The controller minimizes the conditional expectation of the sum of a quadratic position (velocity) error and a quadratic force error while satisfying the constraint of the ARX-model with the estimates substituted for the unknown parameters. The basic approach is used to obtain an adaptive controller which operates on the variables expressed in the joint coordinates. Another adaptive controller is similarly determined for the system variables of the Cartesian coordinate system. The adaptive controller for force-path control introduced here has the form similar to that of hybrid force/position controller of [5], but with time varying gains.

A picture of a scene is used to extract information for an adaptive control algorithm. The object of interest is first located by means of a classifier. The position and orientation of the object are determined from a binary picture. The desired path which the gripper of the manipulator is to follow is specified by discrete points, first is the Cartesian and then in the joint coordinate system. The adaptive self-tuning controller is outlined. The controller design is based on time series difference equation model, in which the parameters are estimated on-line. The gains of the controller are tuned so that a quadratic performance criteria is minimized. The performance of the system designed is then tested experimentally. Results of these experiments are presented.

Power system operation is outlined and system security is defined. The need for security evaluation in power system operation is discussed, and the evaluation is presented as a pattern recognition problem. A suitable straightforward and quick procedure is used to select a small number of variables as features from a large set of variables which are normally available in power systems. Comparison is made on the security classification performances of a number of different types ot classifiers. The training of classifiers is carried out by a search algorithm which seeks to minimize the number of classification errors. The procedure to determine the security functions (classifiers) is illustrated by an example, and simulation results on the steady-state security of the CIGRE 225-kV system show that the pattern recognition approach to security evaluation is encouraging. Some uses of security functions in the control and operation of power systems are outlined. The effect of many of these uses cannot be achieved as easily by other existing methods. Finally, some of the problems associated with the application of pattern recognition to power system security evaluation are discussed.

Linear discrete-time systems containing state and control variables which are randomly delayed are studied. The original system with randomly delayed variables is transformed to another system in which independent random variables appear as plant parameters. For the reformulated system the optimal control which minimizes a quadratic cost is determined. It is also the optimal control for the original problem. The approach is illustrated by an example which describes a production-inventory system.