Engineering topics
Publications and source records attributed to Reid, Max B..
Networks of higher order have enhanced capabilities to distinguish between different two-dimensional patterns and to recognize those patterns. Also enhanced capabilities to "learn" patterns to be recognized: "trained" with far fewer examples and, therefore, in less time than necessary to train comparable first-order neural networks.
The present invention relates to an optical system for creating a potential field map of a bounded two dimensional region containing a goal location and an arbitrary number of obstacles. The potential field mapping system has an imaging device and a processor. Two image writing modes are used by the imaging device, electron deposition and electron depletion. Patterns written in electron deposition mode appear black and expand. Patterns written in electron depletion mode are sharp and appear white. The generated image represents a robot's workspace. The imaging device under processor control then writes a goal location in the work-space using the electron deposition mode. The black image of the goal expands in the workspace. The processor stores the generated images, and uses them to generate a feedback pattern. The feedback pattern is written in the workspace by the imaging device in the electron deposition mode to enhance the expansion of the original goal pattern. After the feedback pattern is written, an obstacle pattern is written by the imaging device in the electron depletion mode to represent the obstacles in the robot's workspace. The processor compares a stored image to a previously stored image to determine a change therebetween. When no change occurs, the processor averages the stored images to produce the potential field map.
A method and system for performing pattern analysis with a neural network coarse-coding a pattern to be analyzed so as to form a plurality of sub-patterns collectively defined by data. Each of the sub-patterns comprises sets of pattern data. The neural network includes a plurality fields, each field being associated with one of the sub-patterns so as to receive the sub-pattern data therefrom. Training and testing by the neural network then proceeds in the usual way, with one modification: the transfer function thresholds the value obtained from summing the weighted products of each field over all sub-patterns associated with each pattern being analyzed by the system.
A Higher-Order Neural Network (HONN) can be designed to be invariant to geometric transformations such as scale, translation, and in-plane rotation. Invariances are built directly into the architecture of a HONN and do not need to be learned. Thus, for 2D object recognition, the network needs to be trained on just one view of each object class, not numerous scaled, translated, and rotated views. Because the 2D object recognition task is a component of the 3D object recognition task, built-in 2D invariance also decreases the size of the training set required for 3D object recognition. We present results for 2D object recognition both in simulation and within a robotic vision experiment and for 3D object recognition in simulation. We also compare our method to other approaches and show that HONNs have distinct advantages for position, scale, and rotation-invariant object recognition. The major drawback of HONNs is that the size of the input field is limited due to the memory required for the large number of interconnections in a fully connected network. We present partial connectivity strategies and a coarse-coding technique for overcoming this limitation and increasing the input field to that required by practical object recognition problems.
Propagation of an electron beam through a gas creates a secondary electron/ion plasma which can have subsequent deleterious effects on the propagation of the beam. In the case of pulsed electron beams with short micropulse durations, these effects can be greatly reduced through the use of a small doping fraction of an electron attachment gas. We present a model which allows the calculation of the reduction in unbound plasma electron density attainable with a gaseous dielectric dopant. Potential problems with a dopant, including increased ionization, increased scattering, altered refractive index, and dopant saturation and fragmentation, are discussed.
A higher-order neural network (HONN) can be designed to be invariant to changes in scale, translation, and inplane rotation. Invariances are built directly into the architecture of a HONN and do not need to be learned. Consequently, fewer training passes and a smaller training set are required to learn to distinguish between objects. The size of the input field is limited, however, because of the memory required for the large number of interconnections in a fully connected HONN. By coarse coding the input image, the input field size can be increased to allow the larger input scenes required for practical object recognition problems. We describe a coarse coding technique and present simulation results illustrating its usefulness and its limitations. Our simulations show that a third-order neural network can be trained to distinguish between two objects in a 4096 x 4096 pixel input field independent of transformations in translation, in-plane rotation, and scale in less than ten passes through the training set. Furthermore, we empirically determine the limits of the coarse coding technique in the object recognition domain.
An algorithm for the optical computation of potential field maps suitable for mobile robot navigation is described and experimentally produced maps and paths are presented. The parallel analog optical computation employs a two-dimensional spatial light modulator on which an image of the potential field map is generated. Optically calculated fields contain no local minima, tend to produce paths centered in gaps between obstacles, and produce paths which give preference to wide gaps. Calculation of 128 x 128 pixel fields at a few hertz are possible with current technology, and calculation time vs. map size scales favorably in comparison to digital electronic computation.
We describe and present experimental results of the optical calculation of potential field maps suitable for mobile robot navigation. The optical computation employs two write modes of a microchannel spatial light modulator (MSLM). In one mode, written patterns expand spatially, and this characteristic is used to create an extended two dimensional function representing the influence of the goal in a robot's workspace. Distinct obstacle patterns are written in a second, non-expanding, mode. A model of the mechanisms determining MSLM write mode characteristics is developed and used to derive the optical calculation time for full potential field maps. Field calculations at a few hertz are possible with current technology, and calculation time vs. map size scales favorably in comparison to digital electronic computation.
The combination of analog optical processors with digital electronic systems offers the potential of tera-OPS computational performance, while often requiring less power and weight relative to all-digital systems. NASA is working to develop and demonstrate optical processing techniques for on-board, real time science and mission applications. Current research areas and applications under investigation include optical matrix processing for space structure vibration control and the analysis of Space Shuttle Main Engine plume spectra, optical correlation-based autonomous vision for robotic vehicles, analog computation for robotic path planning, free-space optical interconnections for information transfer within digital electronic computers, and multiplexed arrays of fiber optic interferometric sensors for acoustic and vibration measurements.
We demonstrate the use of binary phase-only spatial filters (BPOFs) for tracking target sites through a sequence of gray-scale terrain images. The filters are demonstrated in a closed loop guidance system for a laboratory lander mockup. Images of a 3D terrain board taken by the lander's video camera are preprocessed to produce binary intensity contour maps of a simulated planetary surface. A BPOF is made from a section of the current preprocessed image centered on the exact desired landing site. After the lander has descended to a lower altitude, the BPOF is correlated with a new image. The position of the correlation peak is used in making the next filter and to guide the lander so as to recenter the landing site in the camera's view. We present testbed results of the accuracy with which a site may be tracted from orbit to landing, and the maximum scale, translation, and rotation which can be tolerated between subsequent images. Application of the results to NASA's proposed Mars Rover Sample Return (MRSR) and Mars Environmental Survey (MESUR) missions is discussed.
Experiments demonstrated feasibility of synthetic-discriminant-function filter encoded in binary spatial light modulator. Filter is part of optoelectronic apparatus indicating degree of correlation between input image and any number of training images of same object at various positions and orientations. Used in optical correlator to recognize rotated images, or in tracking system to recognize moving vehicle or other substantially rigid object from any direction.
Binary synthetic discriminant function (BSDF) optical filters which are invariant to scale changes in the target object of more than 50 percent are demonstrated in simulation and experiment. Efficient databases of scale invariant BSDF filters can be designed which discriminate between two very similar objects at any view scaled over a factor of 2 or more. The BSDF technique has considerable advantages over other methods for achieving scale invariant object recognition, as it also allows determination of the object's scale. In addition to scale, the technique can be used to design recognition systems invariant to other geometric distortions.
The optical phase errors introduced into an optical correlator by the input and filter plane magnetooptic spatial light modulators have been studied. The magnitude of these phase errors is measured and characterized, their effects on the correlation results are evaluated, and a means of correction by a design modification of the binary phase-only optical-filter function is presented. The efficacy of the phase-correction technique is quantified and is found to restore the correlation characteristics to those obtained in the absence of errors, to a high degree. The phase errors of other correlator system elements are also discussed and treated in a similar fashion.
With the growing interest in using binary phase only filters (BPOF) in optical correlators that are implemented on magnetooptic spatial light modulators, an understanding of the effect of errors in system alignment and optical components is critical in obtaining optimal system performance. We present simulations of optical correlator performance degradation in the presence of eight errors. We break these eight errors into three groups: 1) alignment errors, 2) errors due to a combination of component imperfections and alignment errors, and 3) errors which result solely from non-ideal components. Under the first group, we simulate errors in the distance from the object to the first principle plane of the transform lens, the distance from the second principle plane of the transform lens to the filter plane, and rotational misalignment of the input mask with the filter mask. Next we consider errors which result from a combination of alignment and component imperfections. These include errors in the transform lens, the phase compensation lens, and the inverse Fourier transform lens. Lastly we have the component errors resulting from the choice of spatial light modulator. These include contrast error and phase errors caused by the non-uniform flatness of the masks. The effects of each individual error are discussed, and the result of combining all eight errors under assumptions of reasonable tolerances and system parameters is also presented. Conclusions are drawn as to which tolerances are most critical for optimal system performance.
The state-of-the-art in pattern recognition for such applications as automatic target recognition and industrial robotic vision relies on digital image processing. We present a higher-order neural network model and software which performs the complete feature extraction-pattern classification paradigm required for automatic pattern recognition. Using a third-order neural network, we demonstrate complete, 100 percent accurate invariance to distortions of scale, position, and in-plate rotation. In a higher-order neural network, feature extraction is built into the network, and does not have to be learned. Only the relatively simple classification step must be learned. This is key to achieving very rapid training. The training set is much smaller than with standard neural network software because the higher-order network only has to be shown one view of each object to be learned, not every possible view. The software and graphical user interface run on any Sun workstation. Results of the use of the neural software in autonomous robotic vision systems are presented. Such a system could have extensive application in robotic manufacturing.
A free electron laser is proposed using a periodic dielectric and helical magnetic field. Periodic synchronism between the electrons and the optical wave is obtained at the period of the dielectric and not at the period of the helical magnetic field. The synchronism condition and the gain of the new device are derived. The effects on the gain from dephasing and beam expansion due to elastic scattering of the electrons in the periodic medium are included in the gain calculation. Examples of the resonance transition radiation laser and klystron are given. Operation at photon energies between 2.5 and 3.5 keV with net gain up to 12 percent is feasible using high electron-beam energies of 3 and 5 GeV. Moderate (300-MeV) beam energy allows operation between 80 to 110 eV with up to 57 percent net gain using a klystron design. In both cases, rapid foil heating may limit operation to a single electron-beam pulse.
Optical computing research at NASA Ames Research Center seeks to utilize the capability of analog optical processing, involving free-space propagation between components, to produce natural implementations of algorithms requiring large degrees of parallel computation. Potential applications being investigated include robotic vision, planetary lander guidance, aircraft engine exhaust analysis, analysis of remote sensing satellite multispectral images, control of space structures, and autonomous aircraft inspection.
We address the problem of optical phase errors in an optical correlator introduced by the input and filter plane spatial light modulators. Specifically, we study a laboratory correlator with magnetooptic spatial light modulator (MOSLM) devices. We measure and characterize the phase errors, analyze their effects on the correlation process, and discuss a means of correction through a design modification of the binary phase-only optical filter function. The phase correction technique is found to produce correlation results close to those of an error-free correlator.