Leonard P. Wesley

Planning and reacting in uncertain and dynamic environments

Abstract Agents situated in dynamic and uncertain environments require several capabilities for successful operation. Such agents must monitor the world and respond appropriately to important events. The agents should be able to accept goals, synthesize complex plans for achieving those goals, and execute the plans while continuing to be responsive to changes in the world. As events render some current activities obsolete, the agents should be able to modify their plans while continuing activities unaffected by those events. The Cypress system is a domain-independent framework for defining persistent agents with this full range of behaviour. Cypress has been used for several demanding applications, including military operations, real-time tracking, and fault diagnosis.

Perception-Based Self-Localization Using Fuzzy Locations

We describe a fuzzy-based approach to self localization to support indoor robot navigation. Our approach is perception-based: clues extracted by the perceptual apparatus are matched against an approximate map to obtain an estimate of the robots location in the map. Each perceptual clue is treated as a source of partial locational information, represented by a fuzzy set; other sources, like odometry or external measurements, are also treated in this way. Information coming from different sources is combined using a fuzzy aggregation operator. We illustrate our approach by showing experiments performed on a mobile robot, Flakey.

Evidential knowledge-based computer vision

It has been argued that knowledge-based systems (KBSs) must reason from evidential information, i.e., from information that is to some degree uncertain, imprecise, and occasionally inaccurate. This is no less true of KBSs that operate in the domain of computer-based image interpretation. Recent research has suggested that the work of Dempster and Shafer (DS) provides a viable alternative to Bayesian-based techniques for reasoning from evidential information. In this paper, we discuss some differences between the DS theory and some popular Bayesian-based approaches to effecting the reasoning task. We then discuss some work on integrating the DS theory into a knowledge-based high-level computer vision system in order to examine various aspects of this new technology that have not been explored to date. Results from a large number of image interpretation experiments are presented. These results suggest that a KBSs performance improves substantially when it exploits various features of the DS theory that are not readily available in pure Bayesian-based approaches.

A PROCEDURAL REASONING EXPERT SYSTEM FOR CFD

The use of CFD in the design and analysis of flight vehicles requires specialized knowledge about factors such as algorithm options, gridding methods, physical models, and runtime performance. This knowledge is needed to properly set up a calculation and to monitor the solution during execution of the code. The objective of this work is to develop an intelligent software environment that will assist CFD users of all ability levels in setting up and running CFD codes. Procedural reasoning system technology is well-suited for this task, and a prototype system is used to demonstrate various design concepts. The features of the prototype system include an Input Specification Manager for providing guidance during the initialization stage of a calculation, and a Runtime Manager for monitoring and controlling a running CFD calculation. A library of communication modules is used to easily link the expert system with any CFD code. The goal of this system is to facilitate the CFD design/analysis cycle by minimizing the risks associated with improper input parameters, assisting in the early detection of errors, reducing the need for labor intensive manual checks during execution, and improving user training.

Autonomous locative reasoning: an evidential approach

The authors describe efforts to develop an evidential approach to autonomous locative reasoning that shows promise of overcoming the computational burden of using Cartesian maps, and reduces the need to instrument environments. Perceptual cues were extracted from sonar and structured-light data and they were interpreted to infer the presence or absence of objects found local to the mobile platform. These objects, their spatial relationships, and knowledge of the previously known location were then used to infer the current location. The results of the experiments conducted so far suggest that evidential based locative reasoning techniques can help to bridge the technological gap left by current approaches.<<ETX>>

Evidence-based Power Management in Distributed Sensor Networks

A novel evidence-based approach to energy management in distributed sensor networks (DSNs) is presented. Current approaches and algorithms to DSN energy management are “bottom up” in that they do not consider aspects of the domain of application and particular geographic and environmental situational dynamics to manage energy usage as efficiently as possible. Results from an evidential approach to DSN energy management in a forest fire monitoring and management application are presented. A body of mathematics called evidential reasoning is used to represent and integrate information about the situational dynamics and forest fire detection and monitoring application to draw inferences about the state of nodes in a distributed network. An 18% improvement in energy efficiency compared to the LEACH-C algorithm was observed when the state of the forest environment and knowledge about the dynamics of forest fires are considered managing the state and activity of the DSN. Simulation results using NS2 suggest that the useful life of DSNs can be extended significantly if the situational dynamics and aspects of the domain of application are taken into account.

An automated chest pain diagnostic system for distant space travel

We describe an automated system called Heartache that can take as input diagnostic information such as the location of chest pain, duration of pain, origin and radiation of the pain, and other relevant information to differentiate between possible diagnoses such as non-heart related, anxiety, pericarditis, new myocardial infarction, neuromuscular/skeletal pain, unstable angina, pulmonary embolism, pneumonitis, pneumothorax, stable angina, esophagitis, and tamponade. We also describe how the belief function calculus (a generalization of traditional probabilistic methods) is used to infer the likely pathology from diagnostic inputs that are imprecise, incomplete, and inaccurate to varying degrees. We present the results from using the system on two-hundred diverse medical cases. The system correctly diagnosed all two-hundred cases and was able to explain how the diagnoses was derived, and the factors contributing to the systems conclusions. A successful Heartache system will reduce the need for human medical expertise during distant space travel.

Dynamic risk assessment of CFD codes [aerospace design]

The results of an innovative effort to dynamically extract risk measures that can be factored into the knowledge-base closed-loop control of flow calculations is presented. Example control decisions include changing flow-solving parameters or restarting from previous checkpoints to reduce the need for manual intervention. One significance of the work is a formal, rigorous, yet practical KB means to dynamically control the execution of CFD codes by interpreting solution metrics within the context of larger project-related factors. Such factors include the desired solution fidelity, resource limitations such as budget and time, and related previously completed computational fluid dynamics (CFD) analyses. A measure of the risk associated with continuing, changing, or stopping flow calculations is developed in a manner that can be considered when deliberating if, how, and when to change flow calculations. We demonstrate the benefits of extracting and factoring in measures of risk when controlling the execution of CFD codes.

Dynamic control of the WIND CFD code

Obtaining robust solutions from computational fluid dynamics (CFD) computations remains a topic of interest to the aerospace community. Typically, multiple analysis runs that require manual monitoring and intervention must be done to obtain acceptable solutions. This paper presents the results of efforts to evaluate a particular knowledge-based approach, called the runtime-manager (RTM), to reduce the need to manually monitor and intervene in running CFD codes, and to reduce the computational time needed to reach acceptable solutions. We also characterize the performance of the RTM in a way that allows CFD engineers to predict the potential benefits of using the RTM. Preliminary results suggests that the RTM can help reduce the number of labor hours to complete moderately complex analyses up to 75%, and help reduce computational time up to 44%. We describe the canonical tests and the relatively simple modifications made to the WIND code to achieve these results.

A Real Time Object-Oriented Rational Agent Development System

A real-time object-oriented agent development system (ROADS) is being developed and used to build embedded distributed rational agents in a manner that is difficult or not possible to build with existing agent development environments(ADEs). Some ROADS innovations include the use of a theory of objects as a foundation on which; (1)different agent development languages can be defined and used within a single ADE; (2)computational models of beliefs desires and intentions can be implemented (3)possible world semantics are readily supported, (4)reactivity and responsiveness are under direct dynamic control of the agent applications; and (5)a formal foundation to model cooperative multi-agent applications is provided. ROADS has been successfully used in several complex applications such as distance learning, high-level telecommunications network management, and eventing systems.