ࡱ > f h e [ 6E bjbj "` j j 6A l t t t t t t t 0 L 1 f
0 0 0 0 0 0 0 $ 3 45 0 t 0 t t
0 x N t
t
. 0 o Q- t t .
| 8L@ 3 - . N1 ` 1 . 6 6 . t t t t Knowledge Representation and Ontology for Autonomous Systems
A Proposal for a Workshop at the 2004 AAAI Spring Symposium
Brief Problem Statement and Objective
For an autonomous system to behave appropriately in an uncertain environment, many researchers and practitioners feel that the system must have an internal representation (world model) of what it feels and experiences as it perceives entities, events, and situations in the world. It must have an internal model that captures the richness of what it knows and learns, and a mechanism for computing values and priorities that enables it to decide what it wishes to do. ADDIN REFMGR.CITE Albus2001377Engineering of MindBook, Whole377Engineering of MindAlbus,J.Meystel,A.2001IntelligentIn FileJohn Wiley & Sons, Inc.Wilet Series on Intelligent SystemsAlbus,J.Meystel,A.Zadeh,L.A.2[1]
Autonomous systems in this context refer to embodied intelligent systems that can operate fairly independently from human supervision. A major challenge in autonomous systems is the ability to maintain an accurate internal representation of pertinent information about the environment in which it operates. The inability to do this well hinders effective task planning and execution.
A large body of work exists in various knowledge representation, ontology, and data fusion areas, yet relatively little has been applied to the area of world modeling in autonomous systems. The field of autonomous systems has reached a level of maturity such that it could greatly benefit from leveraging the work that has been on-going in these areas. World modeling in autonomous systems can also serve as a prime problem domain in which to apply theoretical and practical knowledge representation, ontological, and data fusion techniques.
The objective of this workshop is to bring together colleagues in the autonomous systems, knowledge representations, ontology, and data fusion communities to find ways of leveraging existing knowledge technologies to benefit autonomous systems.
Background
There are many types of knowledge that a system must internally model in order to be able to function autonomously. Some of these include:
World Model Data
A priori knowledge: This is often knowledge that is pre-programmed or referenced for an outside knowledge base. This could include maps of the environment, driving rules of the road, road network information, or characteristics of objects expected to be seen in the environment.
In situ knowledge: This is knowledge that is learned or acquired in real-time. The source of this knowledge is based upon information perceived from systems sensors, and then processed to infer information. This could include information about instances of objects in the environment and trajectories of moving objects. a priori and derived from sensor inputs.
Value Judgments - These are representations of the information that can be inferred about the state of the world relative to the task that is being performed. For example, one may infer that it is unsafe to pass another vehicle on a two-lane undivided highway based upon the existence of a solid double yellow line or a vehicle approaching from the other direction.
Gerunds - These are the representations of the present activities that the control system is in the process of doing the yet-to-be-completed output actions (e.g., a vehicle is passing another car, avoiding an obstacle or trying to avoid an obstacle etc.) These provide a set of real-time self-awareness representations for internal reasoning about what the system itself is trying to do.
Representation of the history of task activities The history of an activity could include sequence and timing information, the state of the world at the time activities occurred, decision and value judgments that were made during the course of the activities, and self-awareness states at each point in the process.
Task knowledge These are sets of rules that describe:
how the value judgments are to be derived
how the value judgments and world model data are to be reasoned about to generate the task outputs
how the self-awareness values are to be generated and reasoned about to affect the task outputs
how the history is to be represented and reasoned about to affect the task outputs
Deep knowledge / Knowledge design intent This focuses on the question of Why. Why were certain pieces of knowledge captured as opposed to other? Why was this rule created to determine which action an autonomous system should take? Why is this value used to determine a threshold of when an object is safe to traverse over? This, in some ways, can be seen at meta-knowledge, providing additional information about they knowledge that is stored in the world model.
The knowledge in the internal representation must be modeled in such a format that the autonomous system can make maximum use of it. Different types of knowledge inherently require different forms of knowledge representation. At least three different levels of knowledge are required to be represented and reasoned over. They are: ADDIN REFMGR.CITE Evans2002362Knowledge Engineering for Real Time ControlConference Proceeding362Knowledge Engineering for Real Time ControlEvans,J.Messina,E.Albus,J.Schlenoff,C.2002/9Knowledge EngineeringReal-TimeIntelligentIn FileCrema, ItalyProceedings of the International Workshop on Intelligent Knowledge Management Techniques (I-KOMAT 2002)12[3]
Parametric knowledge: This could include sensory signals, state variables, and system parameters. This type of knowledge is needed to provide position and/or velocity and/or torque control of each degree of freedom by appropriate voltages sent to a motor or a hydraulic servo valve.
Spatial knowledge: This is sometimes referred to as geometric knowledge, iconic knowledge, metrical maps, or patterns. This knowledge is spatial in nature and can be defined as 2D or 3D array data in which the dimensions of the array correspond to dimensions in physical space. The value of each element of the array may be boolean data or real number data representing a physical property such as light intensity, color, altitude, range, or density. Each element may also contain spatial or temporal gradients of intensity, color, range, or rate of motion. Each element may also contain a pointer to a geometric entity (such as an edge, vertex, surface, or object) to which the pixel belongs. Examples of iconic knowledge include digital terrain maps, sensor images, models of the kinematics of the machines being controlled, and knowledge of the spatial geometry of parts or other objects that are sensed and with which the machine interacts in some way. This is where objects and their relationship in space and time are modeled in such a way as to represent and preserve those spatial and temporal relationships, as in a map, image, or trajectory.
Symbolic knowledge: In this form of representation, knowledge is represented by the use of symbols. Symbols could represent objects (nouns) or actions (verbs), and their pertinent characteristics and relationships. A large body of relevant work exists in knowledge engineering for domains other than autonomous systems, such as formal logic systems or rule based expert systems. Whether the knowledge is represented in terms of mathematical logic, rules, frames, or semantic nets, there is a formal linguistic structure for defining and manipulating and using the knowledge. Ontologies represent a major body of work that could play a significant role in autonomous systems. ADDIN REFMGR.CITE Schlenoff2002335Linking Sensed Images to an Ontology of Obstacles to Aid in Autonomous DrivingConference Proceeding335Linking Sensed Images to an Ontology of Obstacles to Aid in Autonomous DrivingSchlenoff,C.2002ImageObstaclesAutonomousArtificial IntelligenceIn FileProceedings of the 18th National Conference on Artificial Intelligence: Workshop on Ontologies for the Semantic Webfile://C:\My Documents\conference papers\aaai2002\WS1402CSchlenoff.pdffile://C:\My Documents\conference papers\aaai2002\aaai02workshop.doc12Niles2001379Towards a Standard Upper OntologyConference Proceeding379Towards a Standard Upper OntologyNiles,I.Pease,A.2001FormalIn FileProceedings of the 2nd Internal Conference On Formal Ontology in Information Systems (FOIS-2001)http://projects.teknowledge.com/HPKB/Publications/FOIS.pdf12[4,5] Potential benefits ontologies can provide include: reuse and modularity, a centralized approach for representing and reasoning with information about the environment, cheaper and more reliable maintenance, increased flexibility of response for the autonomous vehicle, and it also can extend the range of important questions that can be answered to support navigation planning. ADDIN REFMGR.CITE Uschold2003378Ontologies for World Modeling in Autonomous SystemsConference Proceeding378Ontologies for World Modeling in Autonomous SystemsUschold,M.Provine,R.Smith,S.Schlenoff,C.Balakirsky,S.2003World ModelingAutonomousIn FileSubmitted to the IJCAI'03 Conference: Workshop on Ontologies and Distributed Systems12[6]
Detailed Objectives
The field of autonomous systems is continuing to gain traction both with researchers and practitioners. Funding for research is this areas has continued to grow over the past few years, and recent efforts such as the Armys Future Combat Systems Autonomous Navigation System effort is pushing many theoretical research efforts into practical use. However, much research still needs to be performed in the area of knowledge representation, a vital component of many autonomous systems.
A large body of work exists in efforts to develop knowledge technologies in most, if not all, of the categories of information listed above ADDIN REFMGR.CITE Davis1993316What is in a Knowledge Representation?Journal316What is in a Knowledge Representation?Davis,R.1993Knowledge RepresentationIn FileAI MagazineAI Magazine1[2]. However, relatively little work exists on applying these technologies within autonomous systems. Autonomous systems have a need to internally represent information about the environment, and knowledge technologies provide the tools and approach to allow an autonomous system to do so.
This workshop aims to bring together colleagues in the autonomous systems and knowledge technology communities in order to:
Understand what work is currently ongoing in developing knowledge representations and ontologies for autonomous systems
Explore applying various knowledge representation and ontology techniques in the area of autonomous systems
Explore the role that ontologies can play in autonomous systems
Understand which knowledge techniques work best for which types of challenges in autonomous systems
Explore the requirements that other subsystems place on knowledge representation and ontologies: e.g., sensor systems, learning modules, planners, operator control units.
Understand and formalize the interaction between different types of knowledge representations and ontologies that provide different types of information about the same object or event in the environment
Explore approaches to formalize the autonomous systems internal representation
Explore means of measuring the quality and content of knowledge within an autonomous system
Explore the reusability of knowledge among disparate autonomous systems
Determine mechanisms to ensure a tightly collaboration between colleagues in the autonomous systems and knowledge technology communities
Determine how data fusion technologies (which directly support autonomous system sensing capabilities) can be assisted by the use of knowledge technologies. (See: HYPERLINK "http://www.infofusion.buffalo.edu/conferences_and_workshops/ontology_and_viz_ws/ws_products/ontology_action_plan/Ontology%20Action%20Plan.ppt" http://www.infofusion.buffalo.edu/conferences_and_workshops/ontology_and_viz_ws/ws_products/ontology_action_plan/Ontology%20Action%20Plan.ppt)
Prospective Organizing Committee
Definite:
Craig Schlenoff, National Institute of Standards and Technology (NIST)
Jim Albus, (NIST)
Michael Uschold, The Boeing Company
Otthein Herzog, University of Bremen, Germany
Others are being recruited in the areas of:
Autonomous systems
Data fusion
KR and ontologies
Qualitative and Spatial reasoning
Format
The workshop will begin with 20-minute presentations from colleagues in the autonomous system and knowledge technology communities with significant time after each presentation for question and answer. We will then have a series of break-out sessions focusing on the specific topics areas related to knowledge technologies in autonomous systems.
ADDIN REFMGR.REFLIST References
1. Albus, J. and Meystel, A., Engineering of Mind, John Wiley & Sons, Inc. 2001.
2. Davis, R., "What is in a Knowledge Representation?," AI Magazine, 1993.
3. Evans, J., Messina, E., Albus, J., and Schlenoff, C., "Knowledge Engineering for Real Time Control," Proceedings of the International Workshop on Intelligent Knowledge Management Techniques (I-KOMAT 2002), Crema, Italy, 2002.
4. Niles, I. and Pease, A., "Towards a Standard Upper Ontology," Proceedings of the 2nd Internal Conference On Formal Ontology in Information Systems (FOIS-2001), 2001.
5. Schlenoff, C., "Linking Sensed Images to an Ontology of Obstacles to Aid in Autonomous Driving," Proceedings of the 18th National Conference on Artificial Intelligence: Workshop on Ontologies for the Semantic Web, 2002.
6. Uschold, M., Provine, R., Smith, S., Schlenoff, C., and Balakirsky, S., "Ontologies for World Modeling in Autonomous Systems," Submitted to the IJCAI'03 Conference: Workshop on Ontologies and Distributed Systems, 2003.
> { } z { W X [ \
4 p q t u w # # % % , , , , - - s- - . . 1 1 1 1 1 1 1 4 4 d6 e6 h6 i6 '= (= = Ͱ @ OJ QJ ^J NH OJ QJ ^J j OJ QJ U^J OJ QJ ^J OJ QJ ]OJ QJ 6OJ QJ CJ W `t&