The paper describes a test and validation toolset developed for artificial intelligence programs. The basic premises of this method are: (1) knowledge bases have a strongly declarative character and represent mostly structural information about different domains, (2) the conditions for integrity, consistency and correctness can be transformed to structural properties of knowledge bases and (3) structural information and structural properties can be uniformly represented by graphs and checked by graph algorithms. The interactive test and validation environment have been implemented on a SUN workstation.

The bottleneck in the process of building the knowledge base of an expert system is the retrieval of the appropriate problem-solving knowledge from the human expert. Methods of knowledge acquisition and representation from the fields of signal processing, pattern recognition and artificial intelligence are considered in this paper. This unified approach will not only accelerate the knowledge acquisition and organization process, but will also formalize and structure the decision making process by reducing the biases of experts. Using this approach, a Knowledge Monitoring Expert System (KNOMES) has been designed to monitor the waveform signals emitting from a material source. The system consists of four primary components, namely; Fact Gathering, Knowledge Base, Knowledge Formalization, and Inference Engine. The fact gathering subsystem 1) collects the transducer(s) emitted signals and extracts a large feature set from them, and 2) collects the a priori real-world knowledge about the source material through an interface monitored by an expert. The facts, a priori real-world knowledge, and the pattern measurements (features) are organized into a knowledge base. The next subsystem formalizes the knowledge into a tree structure using cluster analysis. The tree structure has proven to be an effective method of information organization and statistical pattern recognition. The last subsystem is the Inference Engine whose one of the component primarily classifies the analytical knowledge. This primary classification is done by traversing through the tree and assigning an appropriate class to an unknown input signal. This paper presents the complete design of the proposed system and outlines the implementation details.

Many robot motion planning algorithms treat the environment in which a robot operates as a meaningless world, that is, a two (or three) -dimensional space filled with obstacles described by geometric functions. For example, some formulate the navigation problem as directing the robot from one place in the space to another without concerning themselves with things such as how the robot decides where it wants to go based on higher level considerations. In order to make a robot truly autonomous, it is necessary to incorporate appropriate knowledge representation and reasoning capabilities in the planning framework. A frame-based knowledge representation and reasoning system, WenLy, is presented in this paper. We will examine the design issues of a knowledge system and specific considerations in WenLy. The structure of knowledge and reasoning mechanisms will be given special attention. We will first present our view of knowledge and knowledge representation, which is very important to the formation of our approach. Then, we will introduce the frame-based system WenLy with some discussions in the structure as a reflection of our view of knowledge. Finally, we will examine some features in WenLy (including other knowledge representation systems) and relate them with the development of knowledge.

The goals of a project underway at Kennedy Space Center include the development of a system capable of creating and simulating the execution of launchsite payload processing plans in which experiments are manifested, prepared for launch and integrated into a Shuttle for transporting to the Space Station for operation. The Space Shuttle then lands with a return cargo which is de-integrated for ground processing after landing. During the execution of this procedure it is necessary to track the utilization and location of all resources and hardware items used to implement the processing such as the pressurized laboratory experiment equipment racks. The knowledge base must be easily maintained and extended by the user and the system should provide a simple user interface in which objects and their associated knowledge base provide the basis for definition, operation and user interaction. This paper discusses the design alternatives and the prototype implementation of the Payload Hardware and Inventory Tracking System (PHITS) and demonstrates how intelligent agents can be used to generate mission schedules. Furthermore, it will be shown how through the use of these intelligent agents knowledge becomes separated into small manageable knowledge bases.

This paper describes Harris research into the use of Expert System technology for the management of the Communications and Tracking System for the Space Station. Harris Corporation has developed the CAMERA (Control and Monitoring Equipment Resource Allocation) Expert System Under a NASA contract with JSC to minimize crew workload in managing the communications of the Space Station. It provides for automatic management of communications resources, diagnosis of faults, and reconfiguration to restore communications automatically. The system utilizes a state-of-the-art man-machine interface to allow high level end-to-end service requests. The expert system interprets the requests, determines the equipment required to implement the service, and assigns the appropriate equipment to the service. The expert system then establishes the service automatically at the time requested and monitors the operation of the simulated system to diagnose faults and determine the appropriate procedures to restore the service. A graphical design tool allows the operator to define new services from existing service primitives. Graphical, hierarchical equipment schematics support both the simulation of faults as well as the diagnostic process. Symbolic models for the equipment and measurements are represented in an object-oriented manner.

Many applications of expert systems to Space Station Automation, such as monitoring, planning, and scheduling will involve reasoning about attributes of objects at different times. For example, in monitoring, the system must reason about changes in signal parameters over time because causal relationships among events are important. In order to reason efficiently and concurrently about attributes with different values at different times, different time formats, and different time validity conditions requires more complex knowledge representations than are generally available in expert systems. Representation issues dealing with point times, intervals, and relative times must also be resolved. We have implemented a temporal reasoning capability in a generic expert system shell (LES) to address these issues and to increase the flexibility of the knowledge representation for a variety of applications. For its first application, we chose monitoring of telemetry data from a satellite (the Space Telescope). Our work involved just the RCE (Rotor Controlled Electronics) bearing, a component of the reaction-wheels subsystem which has attributes such as ACTUAL-TEMPERATURE of the bearing, WHEEL-SPEED, and MOTOR-CURRENT. This task consists of collecting one attribute value per sensor per cycle, checking each value to see if it is within the acceptable range, and storing the each value with a time tag in the database. Processing becomes more complex when one or more readings are out of their acceptable range. The analysis to discover the cause involves examining several cycles of readings, as well as comparing the readings of different sensors over time. The temporal reasoning capability in LES allowed us to compare the most recent readings of two sensors; or to compare one current reading with a value collected some time earlier; or to collect several consecutive readings which are analyzed for trends. In addition, having time tags associated with attribute values permitted us to diagnose different problems occurring at different times with the same component. This would be very difficult to handle without temporal reasoning.

The potential role of automation and robotics (A&R) technologies relating to system autonomy for the Space Station Program has been addressed and evaluated in numerous reports and studies. Of particular importance is the nearly unanimous conclusion that A&R must play a significant role in the achievement of long-term goals for the Station. There are many interrelated issues associated with automating the Station. One major concern is planning for an evolving Station which progressively attains higher levels of autonomy over its lifetime. Planning for an evolving Station is tantamount to planning for continual change. Effective planning for evolution requires that the Station be designed initially in order to accept more than just new equipment and software. This paper addresses some of the more critical hooks and scars which must be incorporated into the Initial Operational Capability (IOC) Space Station in order to pave the way for graceful acceptance of emerging automation technologies in the future.

With the complexities that are likely to be found in control processes aboard the Space Station, it is appropriate to identify a tool for the analysis of these structures at an early stage in the design. The interest in Petri nets for the representation of concurrent structures has grown considerably in the past few years. Introduced here is a robust notation for the machine representation of Petri nets. The notation has been used successfully in the Control Flow and Data Flow representations of systems.

Maintenance of proper functioning of the Space Station will require monitoring of a large number of sensors. This task will include not only state monitoring, but also the need to recognize trends which might lead to fault states. Both types of monitoring would be aided if groups of sensor values could be reduced to a single value which preserved their important characteristics. Multi-dimensional scaling is proposed as a technique to achieve such a goal. This approach, in addition to being useful in the creation of aids to a human operator, would also have characteristics which would make it a useful sensor integration approach for automated systems.

In order for robots to perform meaningful operations in space, some type of sophisticated vision system is required. Such a system should provide the robot with three dimensional (3D) information about its surrounding work space and the objects to be operated on. The problem of remotely determining an object's 3D shape is difficult in any situation, but the space environment presents special problems due to the ambient illumination and lack of atmosphere. The most common imaging systems, visual sensors, are able to accurately determine the boundary of an object's two-dimensional projection onto the image plane. The exact shape of an object can be determined through the analysis of shading in an optical image, but frequently, shading details are obscured by the high intensity specular reflections which occur in space images. An alternate sensor for robotic applications is microwave radar. Some 3D information is available from time-domain radar imaging, but high resolution radar is prohibitively expensive and complex. We would like to integrate the information available from T .V . images and low-resolution radar scattering cross-sections to reconstruct an object's 3D shape. We present a new system for the fusion of optical image data and polarized radar scattering cross-sections. The radar data is used in an iterative procedure which generates successive approximations to the target shape by minimizing the error between the computed scattering cross-sections, and the observed radar returns. The image data is incorporated by supplying an initial estimate of shape through knowledge of the two-dimensional silhouette and shading models. These components are assembled into a larger iterative process designed to refine the estimate of 3D shape and obtain the best possible description of the attitude and motion of the target.

A reconfigurable nonlinear control system design methodology is proposed in this report, to automatically correct computed slew torque commands of space-based pointing systems, for the effects of slew-induced structural deformations. The torque corrections can be generated either from structural sensors, or else from higher derivatives of commanded boresight angular rates. In the latter case, structural control forces must be likewise generated from higher angular slew rates, to shape the slew-induced structural excitation. Automatic selection of slew torque. and structural actuator correction signals, as well as automatic correction for tracking error gains can be managed by a supervisory controller that tests required effort levels against actuator ratings. It is also indicated how translations during slews can be likewise corrected. The simulated implementation of such a control system for a space-based laser beam expander has been implemented on a Silicon Graphics IRIS computer system. A videotape of such a simulation as well as hardcopy are available.

An autonomous intelligent training system has been produced for use by Mission Control Center (MCC) Flight Dynamics Officers (FDOs) training to perform payload-assist module (PAM) deploys from the orbiter. The system (designated PD/ICAT for PAM Deploy Intelligent computer-Aided Training) integrates expert system technology with teaching/training methodologies. Five basic components comprise the system. A domain expert (DeplEx for Deploy Expert) contains the rules and procedural knowledge needed by the FDOs in carrying out a PAM deploy. DeplEx also contains "mat-rules" which encode the errors commonly made by novice FDOs. A trainee model is developed for each individual using the system. The model uniquely represents each trainee's current skill level and history o f interactions with the system. The training system manager (TSM) examines the actions of the trainee and compares them with the actions of DeplEx within the same operational environment. The actions (correct, optional, or incorrect) are noted and the probable cause of incorrect actions is diagnosed using Deplex's mal-rules. A unique feature of the TSM is its ability to permit the trainee the freedom to follow any valid path between two steps of the deploy process. Following each trainee action, evaluative assertions are made by the ITS and used to update the trainee model. A training scenario generator designs appropriate training exercises for each trainee based on his model and the training goal(s) established by the TSM. A user interface has been designed which permits the trainee to access data and take actions in much the same manner as he would at his console in the MCC.

A unique set of problems will encountered in the development of telerobotic systems for space applications such as the Flight Telerobotic System. The dexterous manipulation of objects in zero g will be significantly different. Issues arise from mechanical response and operator interaction with the controls and displays. To reduce development risk, a series of experiments are conceived for the Space Shuttle.

This paper presents a methodological approach to the dynamic allocation of tasks in a man-machine symbiotic system in the context of dexterous manipulation and teleoperation. This paper addresses symbiosis containing two partners that work toward controlling a single manipulator arm for the execution of a series of sequential manipulation subtasks. The proposed automated task allocator uses knowledge about the allocation policies of the problem, the available resources, and the tasks to be performed to dynamically allocate tasks to the man and the machine.

The NASA/NBS Standard Reference Model (NASREM) Telerobot Control System Architecture defines a logical computing architecture for robotics. The architecture provides a framework for integrating a variety of control techniques, and for combining teleoperation and autonomy in one system. This paper demonstrates these aspects of NASREM for the lowest level of the architecture, the Servo Level. The Servo Level supports algorithms for robot manipulator control found in the literature.

Space telerobotic systems will perform complex tasks of assembly, disassembly, and repair of space-based equipment. Planning such tasks requires reasoning about the functional, physical, and geometrical properties of the equipment, as well as a representation of the characteristics and capabilities of the manipulators and sensors available for the task. The AND/OR graph [1,2] is a useful approach to representation of feasible assembly/disassembly sequences and provides the basis for search among alternative strategies. In this paper, we describe the use of parts entropy measures as evaluation criteria for search in the AND/OR graph space. This approach leads to candidate task plans which minimize the complexity of intermediate geometrical states.

Range imagery from a laser scanner developed at ERIM can be used to provide sufficient information for docking and obstacle avoidance procedures to be performed automatically. Three dimensional model-based computer vision algorithms in development at ERIM can perform these tasks even with targets which may not be cooperative (that is, objects without special targets or markers to provide unambiguous location points). Roll, Pitch, and Yaw of vehicle can be taken into account as image scanning takes place, so that these can be corrected when the image is converted from egocentric to world coordinates. Other attributes of the sensor, such as the registered reflectance and texture channels, provide additional data sources for algorithm robustness.

This paper describes a computational architecture for an interconnected high speed distributed computing system for generalized bilateral control of robot arms. The key method of the architecture is the use of fully synchronized, interrupt driven software. Since an objective of the development is to utilize the processing resources efficiently, the synchronization is done in the hardware level to reduce system software overhead. The architecture also achieves a balanced load on the communication channel. The paper also describes some architectural relations to trading or sharing manual and automatic control.

The paper describes the design of a controller for cooperative robots being designed at McGill University in a collaborative effort with the Jet Propulsion Laboratory. The first part of the paper discusses the background and motivation for multiple arm control. Then, a set of programming primitives, which are based on the RCCL system and which permit a program-mer to specify cooperative tasks are described. The first group of primitives are motion primitives which specify asynchronous motions, master/slave motions. arid cooperative motions. In the context of cooperative robots, trajectory generation issues will be discussed and our implementation described. A second set of primitives provides for the specification of spatial relationships. The relations between programming and control in the case of multiple robot are examined. Finally, the paper describes the allocation of various tasks among a set of microprocessors sharing a common bus.

Fuzzy functions, a major key to inexact reasoning, are described as they are applied to the fuzzification of robot co-ordinate systems. Linguistic-variables, a means of labelling ranges in fuzzy sets, are used as computationally pragmatic means of representing spatialization metaphors, themselves an extraordinarily rich basis for understanding concepts in orientational terms. Complex plans may be abstracted and simplified in a system which promotes conceptual planning by means of the orientational representation.

Fuzzy functions, a major key to inexact reasoning, are described as they are applied to planning/grasping. Fractals, in the form of Mandelbrot's Brown line-to-line function are used along with fuzzy functions to describe smoothness, etc. A discussion of knowledge-based systems and common sense includes representation of concepts from the naive physics. Knowledge-based inexact reasoning is used to derive common sense plans based on principles from naive physics.

The design of the Space Station Platforms incorporates and expands on the serviceability and modularity features of the Multi-mission Modular Spacecraft (MMS). This design is adaptable to a wide range of mission requirements through the use of distributed systems with sizing granularity applicable to both multi-use missions, such as earth observations, and large facility missions such as astrophysics observatories. The design permits on-orbit growth, repair, instrument change-out, and resupply of propulsion fuel and cryogen at the Space Station Servicing Facility or with the Shuttle and Orbital Maneuvering Vehicle (OMV). The Platform Program encompasses the design and development of the orbiting unmanned elements of the Space Station to conduct long-term, autonomous, commercial, scien-tific, and technology ventures and investigations. These unmanned elements fall into two general classes of platforms: the Co-Orbiting Platform (COP) and the Polar Orbiting Platform (POP). The COP for the initial operating phase will be developed to support solar and stellar viewing astrophysics missions. The POP will be developed to accommodate earth observation, oceanographic, atmospheric, and solar and plasma physics missions. Current Shuttle payload capability limitations at the Western Test Range (WTR) place a restriction on another objective of platform designs for polar applications, which is to accomplish meaningful science or payload operations after one launch. For this reason, the POP will also be designed to be launched using Expendable Launch Vehicles (ELV's). Features of the Space Station Co-Orbiting and Polar Platforms are described that will allow them to be con-figured optimally to meet mission requirements and to be assembled, serviced, and modified on-orbit.

The Space Systems Integration and Operations Research Applications (SIORA) Program was initiated in late 1986 as a cooperative applications research effort between Stanford University, NASA Kennedy Space Center (KSC), and Lockheed Space Operations Company (LSOC). One of the major initial SIORA tasks was the application of automation and robotics technology to all aspects of the Shuttle tile processing and inspection system. This effort has adobted a systems engineering approach consisting of an integrated set of rapid prototyping testbeds in which a government/university/industry team of users, technologists, and engineers test and evaluate new concepts and technologies within the operational world of Shuttle. These integrated testbeds include speech recognition and synthesis, laser imaging inspection systems, distributed Ada programming environments, distributed relational database architectures, distributed computer network architectures, multi-media workbenches, and human factors considerations.

The evolution of the Space Station's capabilities for customer servicing has been driven by the need to accommodate as broad a set of requirements as possible. At the same time, cost constraints must be weighed against these requirements in order to achieve an affordable program. A thorough analysis of all requirements during the recently completed definition phase of the Space Station Program has led to certain aspects of the Servicing System which will have significant levels of automation associated with them. The key factors which drive these systems in the direction of increasing automation are the limitations inherent in the performance of extravehicular activity (EVA) by the Space Station crew. The Servicing Facility, for example, will incorporate a high degree of automation and teleoperation in its elements in order to free the crew from the burdens associated with EVA. The ultimate goal is to develop the elements of the Servicing System in such a way as to be compatible with and complementary to the Flight Tele-robotic Servicer (FTS). The FTS, being developed by NASA's Goddard Space Flight Center, will evolve to increasing levels of autonomy to allow the virtual elimination of routine EVA. This paper will focus on those aspects of the Servicing System that will incorporate a significant level of automation and the related technology issues that will need to be more fully explored in the upcoming development phase of the Space Station Program.

The Space Station Flight Telerobotic Servicer (FTS) is a flight robotic system for use on the first Space Station launch. The FTS is being designed as a multipurpose tool with the major objective of providing an alternative to astronaut extravehicular activity (EVA) for Space Station assembly, maintenance, servicing, and inspection. Even though neither the Space Station nor the Space Station payloads have been finalized, development schedules for the FTS call for the development of a rational and comprehensive set of FTS functional requirements so that the program can proceed. This paper describes the FTS system architecture and the analysis procedures that are being applied to develop the FTS functional requirements. The FTS Project has adopted a generic, hierarchical control system architecture which allows for comparison of the various potential control system approaches, accommodates potential hardware and software interface requirements, and supports both system enhancements and growth. A Robotic Assessment Test Set (RATS), compatible with the selected FTS system architecture, has been developed to provide a representative sample of potential FTS activities. The intent of the RATS is to help determine the "tall poles" in the FTS design requirements and to provide a basis for comparison of competing FTS design approaches. Each of the tasks in the RATS must stand alone as a mission and has thee following two characteristics: first, the activity concepts are sufficiently advanced that drawings and dimensions are available; and second, the activities/tasks are representative of many of the real tasks that will be performed on the Space Station. Currently thirteen tasks have been analyzed. These RATS tasks involve assembly, maintenance and servicing activities. This paper will describe the FTS system architecture format and the procedure that is being implemented to develop functional requirements for the FTS.

A goal of the telescience concept is to allow scientists to use remotely located instruments as they would in their laboratory. Another goal is to increase reliability and scientific return of these instruments. In this paper we discuss the role of transparent software tools in development, integration, and postlaunch environments to achieve hands on access to the instrument. The use of transparent tools helps to reduce the parallel development of capability and to assure that valuable pre-launch experience is not lost in the operations phase. We also discuss the use of simulation as a rapid prototyping technique. Rapid prototyping provides a cost-effective means of using an iterative approach to instrument design. By allowing inexpensive produc-tion of testbeds, scientists can quickly tune the instrument to produce the desired scientific data. Using portions of the Extreme Ultraviolet Explorer (EUVE) system, we examine some of the results of preliminary tests in the use of simulation and tran-sparent tools. Additionally, we discuss our efforts to upgrade our software "EUVE electronics" simulator to emulate a full instrument, and give the pros and cons of the simulation facilities we have developed.

Satellite diagnosis presents many unusual problems in the application of current knowledge-based diagnosis technology. The operation of satellite systems involves expertise that spans a large variety of systems, hardware, and software design areas. This expertise includes knowledge of design rationale and sensitivities, development history, test methods, test history, fault history and other indications of pedigree, and operational scenarios and environments. We have developed an approach to satellite diagnosis which can integrate evidence from a variety of diagnostic strategies encompassing this expertise. The system utilizes a structural and behavioral model of the satellite, and uses a form of spreading activation to perform the diagnostic procedures on the model. The various sources of diagnostic evidence are combined using a specially-tailored Dempster-Shafer based utility for modelling uncertainty. A prototype of a diagnostic system based on this approach has been implemented.