William H. Brockman

Camera-Space Manipulation

The paper involves the development of real-time estimation and control details for a new vision-based manipulator con trol method. The method permits a kind of adaptability not otherwise available in that the relationship between the cam era-space location of manipulable visual cues and the vector of manipulator joint coordinates is estimated in real time. This is done based upon a model that generalizes known manipulator kinematics to accommodate unknown relative camera position and orientation as well as uncertainty of manipulator grasp. Both large-scale trajectory planning and refined placement precision become possible despite an un known camera-manipulator juxtaposition. This feature opens the door to a range of applications of manipulation, includ ing a mobile manipulator with stationary cameras tracking and providing information for control of the manipulation event. Evidence of the ability of the estimation algorithm to perform in real time is provided with a successfully performed interception task, completed without a priori knowledge of camera or manipulator positions.

Nonholonomic camera-space manipulation

The method of camera-space manipulation is extended to wheeled systems. A minimum of two cameras is required to place points on end-effectors (or objects in their grasp) of n-degree-of-freedom manipulators relative to other bodies where the nonholonomic degrees of freedom on a mobile manipulator base may be included. The target bodies do not have a precisely known location relative to the environment. The method is illustrated experimentally, though not in real time, using a point placement task. It is then generalized to rigid-body positioning tasks. Although the experimental point placement illustrations make use of a very simple trajectory planning scheme for the wheels of the base, a smoother optimal trajectory planning scheme that makes use of the Pontryagin maximum principle is also developed and illustrated. In a departure from the practice of using time as the independent variable for estimation and optimal trajectory planning algorithms, the present development is time independent and instead introduces the forward rotation of the drive wheel of the base as the independent variable. >

Elevation of seizure thresholds: a comparison of cerebellar stimulation, phenobarbital, and diphenylhydantoin.

Generalized EEG seizures were induced in acute, conscious New Zealand albino rabbits with intravenous pentylenetetrazol (PTZ) (10 or 15 mg/kg) or electrical stimulation of the frontal cerebral cortex (ELEC) (50 Hz, 1 msec pulse duration, 2 sec train duration, 4.0–7.6 V). Three anticonvulsant treatments were compared: (1) electrical transhemispheral stimulation of the ansiform or simplex cerebellar lobes (10 Hz, 1.5 msec, 3–4 V), (2) phenobarbital (PB) (25 mg/kg, i.v.), and (3) diphenyl‐hydantoin (DPH) (30 mg/kg, i.v). After treatment, increments in PTZ dose or stimulation voltage were applied until a seizure was evoked that approximated the original in severity and duration. PTZ seizure thresholds were not elevated by DPH, and electrically induced seizure thresholds were not elevated by cerebellar stimulation (CBL). The four remaining seizure threshold elevations (increase in PTZ dose or stimulation voltage) were significant at a level of 0.025 or greater. Comparison of the elevations of seizure thresholds showed no differences at the 0.01 level of significance. Thus, no differences were seen between elevation of PTZ seizure thresholds by CBL or PB, or elevation of electrically induced seizure thresholds by PB or DPH, when examined on an acute basis.

A Simple Electronic Neuron Model Incorporating Both Active and Passive Responses

An electronic model representing a small patch of neural membrane is described. The patch circuit is incorporated into a discrete element ladder network axon model. Electrotonic spread of subthreshold potentials, active action potential propagation, and synaptic events are demonstrated.

A modified cable model for neuron processes with non-constant diameters

Abstract The neuron axon cable model is expanded and developed to describe the linear subthreshold transmembrane potential of any circular cross section, thin membrane neuron fiber whose radius can be expressed as an analytic function of position, r ( x ). The transmembrane time constant is shown under the condition of space clamp to be independent of changes in geometry. Three typical neuron geometries are modeled (dendrite-soma, soma-axon, and dendrite-soma-axon) and the solutions to the resulting differential equations are numerically evaluated. The geometry-induced effects ate attributed to changes in current density and physiological correlates of the effects are proposed.

A camera space control system for an automated forklift

Presents experimental results on a method of camera space control applied to a mobile cart with an on-board robot, operated as a forklift. The objective is to extend earlier results to the task of precise and robust three-dimensional object placement. The method is illustrated with a box stacking task. >

Restoration of noisy images blurred by a random medium

An image restoration model is given to restore images blurred by a random medium. The overall distortion process caused by the random medium is represented by a stochastic point spread function. To estimate the original images degraded by this model, two restoration algorithms are formulated based on the Wiener filter and the constrained least-squares filter. Computer simulations are included to compare the performances of the proposed algorithms with the conventional ones.<<ETX>>

Ladder network prediction of transients in linear spatially inhomogeneous cables

Abstract A numerical method is described for finding steady state and transient responses in electrically linear, spatially inhomogeneous cables. Spatial inhomogeneities are incorporated by representing the cable by a number of finite length uniform cylindrical segments, each having the radius and electrical characteristics of a small region along the cable. Input waveforms are approximated by truncated Fourier series of sinusoidal components. Output waveforms are produced by multiplying the input Fourier series sinusoids by their respective transfer functions between input and output points on the cable and summing the resultant output point sinusoids. The transfer functions, representing attenuation and phase shift for each input sinusoid, are obtained by numerical analysis of an electrical ladder network derived from the cylindrical segment model of the cable. Results are shown for application of this method to both cylindrical and expanding radius cable geometries.

Fixed classifier pattern recognition using iteratively produced preprocessing

Pattern recognizers are often composed of two parts, the feature extractor and the classifier. This paper is a description of a pattern recognizer whereby the classifier learns first, and is then fixed, followed by learning by a preprocessor, which must learn how to predistort the input to the fixed classifier for proper recognition of the learning set. Learning the distortion is an iterative process whereby each vector of the training set must be examined for each iteration. Each iteration fixes the parameters for several fundamental distortions, and the use of a subset of all the distortions over all iterations constitutes a net distortion.