Reda E. Fayek

Using a blackboard to integrate multiple activities and achieve strategic reasoning for mobile-robot navigation

We have designed a mobile robot system architecture that uses a blackboard to coordinate and integrate several real time activities. An activity is an organizational unit, or module, designed to perform a specific function, such as traversing a hallway, going down steps, crossing over an open channel on the floor, or tracking a landmark. An activity resembles a behavior in that it controls the robot to perform a specific task. It differs from a behavior in that it is designed to perform the specific task in a narrow application domain, whereas a behavior generally resembles a biological response-that is, an organisms response to a stimulus. The activity based blackboard system consists of two hierarchical layers for strategic and reactive reasoning: a blackboard database to keep track of the state of the world and a set of activities to perform real time navigation. >

Using hypergraph knowledge representation for natural terrain robot navigation and path planning

Rapidly changing requirements in manufacturing and robotics require efficient automated planning systems. In this paper, we present a method to acquire and exploit domain-knowledge. We use two examples of knowledge-extensive contexts; outdoor terrain robot navigation and mission planning. We represent the acquired sensory 3D data by triangular terrain meshes. Application independent features are automatically extracted from these and converted into symbolic entities suitable for reasoning. Their topological relations are then organized into attributed graphs. Higher-order, application dependent relations are captured by hyper-edges in attributed hypergraphs. The symbolic relations inducing hyperedges are used as the basis of symbolic reasoning operations. The resulting compact hypergraph representation of the raw data facilitates complex navigation and mission planning tasks. Domain-knowledge is thus captured in a flexible form and used to reduce the search for feasible paths.

Feature-based 3D surface modeling by improved constrained triangular meshes

Triangular meshes are suitable representations for the large amounts of 3D data produced by range sensors. 3D object recognition tasks are complicated due to the high resolutions involved. We present a new technique to improve the interpretation quality of triangular meshes and to bridge the gap between raw sensory data and symbolic models used for recognition. We efficiently describe the surface features and avoid insignificant high resolution details. We accentuate the surface features by locally and incrementally improving the mesh topographic properties. The resulting locally-optimal triangular mesh removes the artifacts inevitably introduced by scene digitization. This simplifies subsequent object recognition tasks.

Triangular mesh model for natural terrain

Planetary exploration is a new challenge in the field of space robotics. Most of the topographic information available today about planetary surfaces was gathered either by orbiters or stationary landers. The use of an autonomous rover capable of building detailed maps of unstructured natural terrains would be more relevant, in the long term, for such space exploration missions. This paper presents a natural terrain model to represent and analyze the information gathered by an exploration rover using 3D computer vision. Irregular triangular meshes are suitable for the representation of amorphous natural terrain and could easily adapt to its irregular nature. The triangular terrain model presented captures both the range and the intensity information of the planetary scenes, enabling the modeling of topographic features at different resolution levels and the construction of conceptual topologic maps. In this paper, we first provide a compact representation for the rugged terrain scenes, and then use the resulting topographic maps to assist visual navigation of an autonomous exploration rover. The symbolic topologic maps can be used for the strategic planning of the exploration mission and reasoning about its various alternatives.

A Blackboard, Activity-based Control Architecture For A Mobile Platform

This paper presents a system architecture for autonomous navigation for a mobile platform designed on the view of activities. A control unit that acts upon informa- tion posted on a blackboard supervises these activities. Each activity is composed of several sensing modules connected to one or more perception modules, which are, in turn, con- nected to an appropriate action module so that the platform can perform a defined functionality. The control unit inter. acting with the blackboard decides what activity should be active at any given time, based on information provided by the perception modules. Interaction between activities is minimal or none, and, therefore, it is possible to design and debug each activity independently. Information detected by the perception modules is posted on the blackboard to be processed by the control unit. This control unit has the op- tion to alter the parameters of the current activity, or to deactivate its action module so that another activity can be enabled. It acts as a high level reasoning agent, while the activities can perform the desired functionalities of the vehi- cle in real time. This architecture supports the use of multi- ple sensors; they are treated as modules that can easily be connected to the,perception modules. This leads to an activ- ity-based system structure which is highly flexible and adap- tive. Defining the activities in this control architecture is much easier than defining levels of abstraction in a hierar- chical structure and less error-prone than describing layers in a behavior-based controller.

Adaptive MOEMS for Dynamic Beam Focusing

Micro-opto-electro-mechanical systems (MOEMS) are finding increased uses in instrumentation and many imaging applications. The required complex optical manipulations of laser beams use expensive, sub-optimal and inaccurate optical components (lenses, mirrors, spectral filters, etc.). MOEMS components can be introduced instead of (or in addition to) these optical path elements. Such MEMS devices can significantly improve the overall performance of these instruments. We propose MOEMS elements arrays to act as adaptive and corrective optics. Alternatively, the MOEMS elements can dynamically vary the optical path structure. As a proof of concept, a simple 2D array of micro-mirrors that approximate a concave mirror is investigated. CoventorWare 3D modeler was used for the design and simulations. The prototype implementation was done using PolyMUMPs. This dynamic focusing device changes the focal point of an incident laser beam with low actuation voltage and tuning voltages.

Adaptive MOEMS applications in telemetry and multi-spectral imaging

Micro-opto-electro-mechanical systems (MOEMS) are finding increased uses in instrumentation and many imaging applications. The required complex optical manipulations of laser beams use expensive, sub-optimal and inaccurate optical components (lenses, mirrors, spectral filters, etc.). MOEMS components can be introduced instead of (or in addition to) these optical path elements. Such MEMS devices can significantly improve the overall performance of these instruments. We propose MOEMS elements arrays to act as adaptive and corrective optics. Alternatively, the MOEMS elements can dynamically vary the optical path structure. Initial prototypes have been built. Sample applications are introduced. They span multi-spectral medical imaging and laser radar telemetry.

Adaptive MOEMS mirrors for medical imaging

This paper presents micro-electro-mechanical-systems (MEMS) optical elements with high angular deflection arranged in arrays to perform dynamic laser beam focusing and scanning. Each element selectively addresses a portion of the laser beam. These devices are useful in medical and research applications including laser-scanning microscopy, confocal microscopes, and laser capture micro-dissection. Such laser-based imaging and diagnostic instruments involve complex laser beam manipulations. These often require compound lenses and mirrors that introduce misalignment, attenuation, distortion and light scatter. Instead of using expensive spherical and aspherical lenses and/or mirrors for sophisticated laser beam manipulations, we propose scalable adaptive micro-opto-electro-mechanical-systems (MOEMS) arrays to recapture optical performance and compensate for aberrations, distortions and imperfections introduced by inexpensive optics. A high-density array of small, individually addressable, MOEMS elements is similar to a Fresnel mirror. A scalable 2D array of micro-mirrors approximates spherical or arbitrary surface mirrors of different apertures. A proof of concept prototype was built using PolyMUMPTM due to its reliability, low cost and limited post processing requirements. Low-density arrays (2x2 arrays of square elements, 250x250μm each) were designed, fabricated, and tested. Electrostatic comb fingers actuate the edges of the square mirrors with a low actuation voltage of 20 V - 50 V. CoventorWareTM was used for the design, 3D modeling and motion simulations. Initial results are encouraging. The array is adaptive, configurable and scalable with low actuation voltage and a large tuning range. Individual element addressability would allow versatile uses. Future research will increase deflection angles and maximize reflective area.

Extracting buildings from aerial topographic maps

The recovery of 2D information from intensity images, and that of 3D information from range images are the major issues in 3D objects recognition from sensory data. The analysis and interpretation of remote sensing aerial images have important applications. This paper presents an efficient method for the analysis and modeling of such scenes based on range sensory data. Unlike methods using 2D intensity images, we exploit the rich 3D data. We extract symbolic information from the 3D triangular mesh models. These are used to recognize buildings using symbolic reasoning and generic object models.

Using random graphs in learning past tenses

This paper presents an automatic learning technique. Its strength is demonstrated by a bench mark problem for evaluating learning systems-learning the past-tenses of English verbs. To deal with training patterns of variable length and to exploit probabilistic properties inherent in the English language, our method uses random graphs. Due to the large amount of redundancy between the base form and the past tense form of most verbs, only the endings of these forms are of interest in the classification process. These are extracted automatically. Taking advantage of the sequential nature of words, the complete graphs can be relaxed to sequences with great reduction in the computational requirements. A significant merit of this method over the connectionist models is that the knowledge acquired is expressed automatically in a rule-based form which simplifies the classification process of unknown verbs.<<ETX>>