There are three parts to this project. The Double Vision Computing Paradigm for Artificial intelligence; The Double Vision Computing Theory; and the Computational Epistemics for Vision. The present paper is only an overview of part 1 and its applications to autonomous spacecraft vision.The other two parts are to be projects are being defined by the papers referenced. First the optimistic AI design paradigm is presented from our earlier papers and it is shown how it is transformed to a new computing paradigm, with analogies to computing vision. Designs with object coobject pairs is presented for defining Double Vision Computing.

Linguistics definitions for intelligent languages and objects from our papers Nourani (1993,1994) are applied to present the basic technical definition for computing with object co-object pairs which defines double vision computing syntax and objects A brief view of the forthcoming project that we call. The Computational Epistemics For Vision is put forth to present the computational Cognitive initiative that is implied by part three the present project. The techniques for abstract implementations by computing agents are further presented for implementing Double Vision Computing by AI agents that cooperate n boards for problem solving. The mathematical view for part two of the project, the Double Vision Computing Theory, is outside the intended goals of the present extended abstract and is a subject for forthcoming papers.

Eversince writing (Nourani (1993) we have put forth an AI design technique that is an optimistic problem solving paradigm. The paradigm is to design as if nothing could go wrong, then account for all faulty behavior, recoverable activities, and exceptional behavior by defining preconditions on coobjects, and agents defined on the co-object that can make intelligent recovery from unusual behavior. This techniques promises to make it easier to design AI systems for complex problems with many objects and thousands of rules. It is quite difficult to have general problem solving paradigms that can 'get to the point' when there are so many exceptions and unusual things that are the 'last thing on the mind' for a particular problem solving technique Nilsson (1971).

The methods noted in Nourani (1993) allow us to accomplish this in a systematic way. The reader might ask how such dual conceptualization can be conceived in design. That is a good question to address. The way design is viewed by our approach in Nourani (1993) is that , there are objects actions and relations defining functionality. In the present approach at the knowledge acquisition phase the expert is to state all exceptions to actions and what recovery and corrective actions are to be carried out. Thus for each action on an object a dual action is to be supplied by the knowledge acquisition phase, such that a specifier can fully define the effect of the dual actions.

As an illustration the following trivial example defines part of a flight systems operations. In the example below O is Enterprise_Flight System, Transport an example of an action, a member of A, EXP defines the set F, the third line defines an example of a relation in RLA, and the last few lines are examples of relations in RRA.

Object:= Enterprise_Flight_Computer

OPS:= Transport (Source,Destination,Object)| ......

Transport (Mars,Enterprise, Module) => Signal an available Mars_Rover robot to Fetch Module from <stored_location, module number>;

Set module at Locations Scanned by the Transporter scanning device;

Energize

Exp:= ETransport (Source,Destination,Object) |...

ETransport(Mars, Enterprise,Module) =>

If Mars_Rover robot signals that Module is locked or too heavy for robot to move to transportable points, Activate_Multiple_Robots to cooperate, unlock and move module to transport point.

If robots encounter obstacles:

signal to identify the obstacle;

propose alternate route, and signal obstacle avoidance functions

Compute the source to destination trajectory every fraction of the second;

Signal robots the transporter probe area;

Interrupt Energize function if object to be transported is not within probe range.

In the above example OPS denotes operations, EXP denotes exceptions, and the last equation defines the exception action. In this example there is a process(action) that is always checking for various faults and trajectories and signaling robots. When defining actions on objects the specifier must have a set of preconditions to check before an operation or action is permitted. Such set of preconditions then automatically form a set of processes to check for validity of an operations.

This automatically implies that a dual set of actions are to be defined on the co-objects that can take the failure of a precondition and define the alternate recovery actions or remedial actions that are to take form on the objects. For example, in the above Enterprise example a precondition to Energize is that the module or object to be transported is within the probe range of the transporter. That automatically implies that there should be a dual set of functions defined to deal with the possibility of that the object is not within range.

This duality for a problem solving paradigm is made symmetric by the present paper to formulate Double Vision Computing. The basic technique is that of viewing the world as many possible worlds with agents at each world that compliment one another in problem solving by cooperating. An asymmetric view of the application of this computing paradigm was presented by the author and the basic techniques were proposed for various AI systems Nourani (1993). The double vision computing paradigm with objects and agents might be depicted by the following figure. For computer vision, the duality has obvious anthropomorphic parallels.

The object co-object pairs and agents solve problems on boards by cooperating agents from the pair without splurges across the pairs. The term splurge has a technical definition for object level computing presented in Nourani (1993,1994) and section 3 The cooperative problem solving paradigms have been applied ever since the AI methods put forth by Hays-Roth (1985). Computing by agents might apply the same sort of cooperative problem solving methods. (see figure 1)

Fig 1 The Multiagnet Multiboard Visual Field

This section is written only to present a motivational overview of another research area and paper to be presented elsewhere, and to show that there is a formal basis for the computing paradigm depicted by figure 1. In Nourani (1993,1994) we have advocated programming techniques with intelligent trees and objects, where programming language automatic implementing and semantics might be defined form the abstract syntax trees Nourani (1993c) and language constructs. We have put forth the generalized diagram method to encode the model-theoretic semantics of a language from its abstract syntax. We have presented language designs with linguistics constructs that make it easier to identify G-diagram Nourani (1984, 1987,1980,1991) models and define automatic implementations from abstract syntax Nourani (1993b,f,g).

The computing enterprise requires general techniques of model construction and extension, since it has to accommodate dynamically changing world descriptions and theories. The models to be defined are for complex computing phenomena, for which we define generalized diagrams. These newer techniques for building models for theories and for AI applications has appeared in this authors papers Nourani (1987,1991,1993a,b,c,e,1995b), Nourani-Hoppe (1994) and is referred to by the G-diagram techniques.

The technical examples of algebraic models defined from syntax had appeared in defining initial algebras for equational theories of data types Nourani (1980,1991) and this author's mathematical logic papers Nourani (1993). In such direction for computing models of equational theories of computing problems are presented by a signature defining the names and arity for the functions symbols, and a set of equational axioms.

The generalized diagram (G-diagram) Nourani (1987,1991) is a diagram in which the elements of the structure are all represented by a minimal set of function symbols and constants, such that it is sufficient to define the truth of formulas only for the terms generated by the minimal set of functions and constant symbols. Such assignment implicitly defines the diagram. The minimal set of functions that define a G-diagram are those with which a standard model could be defined.

By an intelligent language we intend a language with syntactic constructs that allow function symbols and corresponding objects, such that the function symbols are implemented by computing agents in the sense defined by this author in Nourani (1993c). A set of function symbols in the language, referred to by Agent Function Set, are function symbols that are modeled in the computing world by AI Agents.The objects, message passing actions, and implementing agents are defined by syntactic constructs, with agents appearing as functions, expressed by an abstract language that is capable of specifying modules, agents, and their communications. We have to put this together with syntactic constructs Nourani (1994) that could run on the tree computing theories Nourani (1993b,1994).

Definition 3.1 We say that a signature is intelligent if it has intelligent function symbols. We say that a language has intelligent syntax iff the syntax is defined on an intelligent signature.

Definition 3.2 A language L is said to be an intelligent language if L is defined from an intelligent syntax.

Definition 3.3 We say that a function f is an information string ,iff there is no message passing or information exchange except onto the object that is at the range set for f, reading parameters visible at each object. Otherwise, f is said to be an information splurge We refer to them by string and splurge functions when there is no ambiguity.

Remark: Nullary functions are strings functions.

The formulation overview presented here form our theoretical papers on computing with intelligent trees and objects Nourani (1993b,c,1994), formalizes the sense in which the double vision computing paradigm in figure 1 is defined. The linguistic tree computing formulation of the present computing techniques might also make it easier to implement tree search algorithms to implement AI problem solving Genesereth and Nilsson (1987). However, that is a topic for another paper.

The present paper and (Nourani 1995 a) only presents a glimpses of what might be the subject of A Computational Epistemics of Vision by Nourani (1997), only to present a technical basis for a view of computer vision computation. To connect the present computing paradigm with computer vision Winston (1975) , to the extent that the present extended abstract could allow, we only propose that vision is by computing agents that cooperate on visual computing, eyes helping focus one another to resolve ambiguities, to say the least. We are not at a point where theoretical advances might be reported yet Nourani (1993a,1995). However, we only state that the analogy to computer vision is not a coincidence.

There is only a glimpse of what this author has termed The Vision Computational Epistemics on informal memos. A theory and techniques for connecting the artificial intelligence computing epistemics Nourani (1993,1994,1995) to the Cognitive part of vision Kirkeboen (1992) might be defined, but only when we have "warp" passed the meteor shower inevitable with the mathematical theory for double vision computing, AI, epistemics Nourani (1994) and Cognitive paradigms that are yet to be discovered. At present we have only scanned for Cognitive computing life and intend to develop this area of research starting from The Computation Epistemics for Vision by (Nourani 1997).

The term "agent" has been recently applied to refer to AI constructs that enable computation on behalf of an AI activity Genesereth and Nilsson (1987). It also refers to computations that take place in an autonomous and continuous fashion, while considered a high-level activity, in the sense that its definition is software/hardware, thus implementation, independent Nourani (1993c).

For example, in a planning Fikes and Nilsson (1971) problem for space exploration, an agent might be assigned by a designed flight system to compute the next docking time and location , with a known orbiting space craft. Agents are in most cases informable, thus allowing message passing actions. We can define Abstract Intelligent Implementation (AII) Nourani (1993b,c) methods as intelligent agent architectures , with external behavior that is a function of the degree of message passing actions and parallelism conceptualized.

Since our problem descriptions consist of objects, actions, and relations defining the effect of actions on objects, it is not difficult to show that they can be implemented by a collection of active agents that communicate through their operations, parameters and messages as conceived by the knowledge acquisition phase.

The specifications <O,A,R>, once expanded further, are in fact of the form <O,(LA,RA),(RLA,RRA>, where LA and RA are the actions, and (RLA,RRA) their respective relations, LA for left-vision and RA for right vision. We had invented Nourani (1992) a necessary twin-turbojet <FLA,FRA>, consisting of FLA := <O,LA,RLA> and FRA;= <O,RA,RRA>. FLA corresponds to an algebra Alg[LA] of actions, with an anthropomorphism to the Left Eye, and FRA to an algebra Alg[RA] for the Right Eye. Each of the FLA and FRA consists of agents that are mutually, often pair-wise, informable. This defines a pair of systems, each consisting of a collection of objects, actions and relations.

Actions could be in form of operations or message communication from one object to another. In other words, a collection of computing agents forms FLA and a dual collection forms FRA. This defines a concurrent collection of systems that can be realized by agents that logically or physically might be thought of as running on several microprocessors.

The degree of precision for computing and vision is a function of the intelligence of the agents implementing the <FLA,FRA> pair. The agents have incomplete information about the immediate needs of activating other agents or exceptions. Thus the efficiency and strength of functionality of our AI systems are a function of the degree of intelligence we build in the implementing agents. The <FLA,FRA> are defined from the knowledge acquisition inputs.

We have applied the techniques for design and implementation of autonomous robot supervision systems. Robot supervision is an area of crucial importance to which multiagent AI techniques might be applied to solve real problems encountered in fields such as intelligent systems, aerospace, robotics, etc. Each of the stages has to be approached in ways that define supervision and recovery from faults. The recovery problems had been studied and are quite important for robot design (Gini's [1983] and Nourani [1993b]). However, since the multiagent AI techniques are recent the present paper on MARS- Multiagent Robot Supervision, is perhaps the first to formally address robot fault and recovery with multiagent AI. System definition is by independent concurrent computing agents.

The present paper defines FTS by a pair of systems, each consisting of many computing agents. The two systems are mutually synchronized to enable fault and exception handling and recovery in an automatic manner. The paper also presents AI techniques for implementing FTS.

The initial phase of designing FTS is to present the design in form of a specification[2]. For the present project, the approach implies starting with supervision knowledge definition phase. The Supervision Knowledge (SKA) is defined at the conceptualization stage. This is a crucial stage to the design for MARS. Thus specification are triples <O,A,R> consisting of objects, actions and relations. Actions are operations or processes. A recent project applying the present techniques is reported by this author in [Nourani 1993a,c]. The problems of abstraction [Nourani 1980,93b,c], object-level programming, and agent view of AI computation are the important components of inter-play in the present paper. This differs from the usual approaches to knowledge acquisition since knowledge acquisition in such approaches merely consists of obtaining knowledge in clausal form relating the expert knowledge to the AI expert, without any structural requirements on the expert knowledge imparted.

SKA has some additional requirements to be put forth. The requirement is that each object to be defined has to have a dual definition in terms of the actions that are taken for exception and recovery. Thus at the knowledge acquisition phase the expert is to state all exceptions to actions and what recovery and corrective actions are to be carried out. Thus for each action on an object a dual action is to be supplied through SKA, such that a specifier can fully define the effect of the dual actions. Thus the supervisory, exception and fault actions are fully specified in each case.

The term "agent" has been recently (see [Genesereth-Nilsson 1987], for example) applied to refer to AI constructs that enable computation on behalf of an AI activity. It also refers to computations that take place in an autonomous and continuous fashion, while considered a high-level activity, in the sense that its definition is software/hardware, thus implementation, independent (Nourani 1980,93c]). Such functionality is typical of what is required to implement autonomous planning systems (Nourani 1991,93c,95c). Agents are in most cases informable, thus allowing message passing actions. Thus cooperating expert systems can be viewed as simple examples for the agent models of AI systems.

We can define a robot supervision systems designed by AI methods as intelligent agent system (Genesereth and Nilsson 1987), with external behavior that is a function of the degree of message passing actions and parallelism conceptualized. The fault tolerant specific architectural issues are dealt with further in the next two sections. Since our specifications consist of objects, actions, and relations defining the effect of actions on objects, it is not difficult to show that they can be implemented by a collection of active agents that communicate through their operations, parameters and messages as conceived by the SKA phase.

The specifications <O,A,R>, once expanded further, are in fact of the form <O,(A,F),(RNA,RSA)>, where A is actions, F is faults, and (RNA,RSA) their respective relations, NA for normal action and SA for supervision action. In the example of the next section many objects and their fault and supervisory functions are presented. Now we present a necessary twin-engine turbojet <FNF,FSF>, consisting of FNF := <O,A,RNA> and FSF;= <O,F,RSA>, for FTCS with AI techniques. Each of the FNF and FSF consists of agents that are mutually, often pair-wise, informable.

<FNF,FSF> defines a pair of systems, each consisting of a collection of objects, actions and relations. Actions could be in form of operations or message communication from one object to another. In other words, a collection of computing agents forms FNF and a dual collection forms FSF. This defines a concurrent collection of systems that can be implemented by agents that logically or physically be thought of as running on several microprocessors.

The degree of fault tolerance is a function of the intelligence of the agents implementing the <FNF,FSF> pair. The agents have incomplete information about the immediate needs of activating other agents or exceptions. Thus the efficiency and fault tolerance and recovery of our software systems are a function of the degree of intelligence we build in the implementing agents. The agents must have some reasoning ability to make transition or message passing decisions.

This approach allows us to design systems that can deal with unplanned or erroneous behavior in an AI system. In figure-2, section 9, the agents ai are implementing functions that are defined on the objects. The next step is defining the <FNF,FSF> from the supervision knowledge definition (SKA) inputs. Its realization consists of a pair of communicating systems to be defined in the following sections (see figures in section 9.)

This approach has a mathematical computing model, to be presented in subsequent future expositions, consisting of an algebra of processes and objects. Preliminary theoretical papers have been written by this author during the current year, supporting the computing techniques applied here.

In this section we take the reader through the design stages of a typical autonomous robot flight system for real applications involving many modules with independent functionality. The method of design and implementation by multiagent AI techniques is illustrated by a flight control example. A typical complex problem domain is the design of autonomous space flight system with supervisory functions.

The expert is to come up with a set of functions, each corresponding to a space flight module, and to define how the modules interact and function together, by defining some operations amongst the modules. The modules are each complex hardware-software systems, best thought of as a microprocessor with its own running software, that implements the functionality of the module as extrapolated from the design process.

Let us view one sample set of modules: M1-M8, each representing a functionality Fi, respectively. The modules are functions defined applying the visual field - figure 1, section 2.

M1: Thrust Control

M2: Stage Control

M3: Orbit Selection

M4 : Attitude Control

M5: Flight Deck Control

M6: Sensors

M7: Obstacle Avoidance

M8 Communications

M9: Docking Functions

We present the implementation of one of the modules, M1, for the reader, the function F1 that is to implement M1. To define the thrust control there are a number of parameters that come to play that are hardware implied data or functionality related requirements. These are to be specified and then implemented by AI agents. Each function defined on the object corresponding to M1 is specified with a set of preconditions that imply co-objects to be defined for exception and recovery if the preconditions fail.

Object:= Thrust Control = <TCN, TCF> corresponding to normal and supervision co-object

Ops:= Throttle Level Up (TLU) | Throttle Level Down (TLD)

Preconditions to be defined are on the following objects or parameter set (PS):

{velocity, acceleration, attitude and tilt level, trajectory, control panel, obstacle encounters}, An operation TLU on the object M1, thrust control = <TCN,TCF>, is preconditioned by the above parameters that are implemented by functions that are defined on the co-object TCF for supervision and to check for supervision, faults and exceptions. For example, when defining TLU the designer must keep in mind a velocity preconditions relative to the various parameters, that must be implemented by a function defined on the co-object.

That function, let us refer to it by velocity check (vc) is a process that always checks the velocity to make sure it does not exceeds limits. Similar SA and fault functions are defined for all the parameters in the set PS. Thus on the object TCN there is an operation TLU, and on the coobject TCF there is an operation vc to implement velocity check preconditions for normal functioning of TLU. Each of the functions defined on the object or co-object are implemented by many agents. For example, when the velocity check function is invoked one or all of the following set of functions get activated.

An agent vca1 is invoked to signal M1 to activate some level of TLD, while checking with agents running off of panel and sensor readers. It also signals flap controllers if the craft is at stage where flaps are effective, to try to recover from hazardous conditions. If all fails it activates agents that are to implement automatic thrust control to bring velocity to acceptable levels.

Similar set of agents are implementing for acceleration, attitude and tilt levels, trajectory violation, and obstacle encounters. In each case the co-object functions and their implementing agents try to compensate such that the precondition to an operation on the object is met.

In the figure below objects are represented as <object,co-object pairs>, where the co-object is a copy of the object on which supervision, faults and recovery functions are defined.

The functions are implemented by agents ai, where ai are agents implementing velocity, acceleration, obstacle avoidance, or agents that check for limitations of the functions.

.

Figure 2 The Thrust Module- The Mi's are object-coobject pairs and the ai's the agents. The dotted lines are agent message paths. The thurst has directional vision and agent controllers off the boards.

The above multi-agent realization of the specifications implies design with a pair of concurrent systems. Each of the two systems is to be designed with a collection of modules, such that there corresponds a module for each specification. A module consist of the minimal set of processes and objects that can be used as a basis for defining a computing activity. This term is analogous to the terminology familiar in operating systems concepts.

The objects and the operations of one set of modules once defined specifies the FNF, while those of the FSF are defined by the dual module. The set of modules defining FNF and FSF are synchronized by cross operations and interact by some operations that are implemented by message communications between FNF and FSF. These operations are defined to either inform the various processes that are mutually dependent or to take the system from an active state in FNF to an active state in FSF. Note that when exceptional conditions occur the active state is FSF. However, both collection of modules are considered concurrently "running."

FSF's major task is supervision, fault handling and recovery. If fault recovery is taking place, in each module, the active module (a collection of agents) for a particular function, will be the FNF component, while the FSF component does concurrent checks for further exceptions should they be encountered. In each of the modules there are objects, processes defining the operations, and objects to which there is a corresponding function in the other module. Thus FNF and FSF are a collection of objects and processes. FNF := <{O1,<p1,...,pn>},{O2,q1,q2,..},...{On,...},RNA>

RNA is the set of relations on each object and cross objects.

FSF := <{O1,<e1,...,en}>,{O2,<e11,e12,...,e1m>},...,{On,<...>},RF>

RSA is the set of relations on each objects and cross objects.

Figure 2- Agents Implementing Functions

Figure 4 Supervision Agents

Each of the processes can have a corresponding agent in the dual family. The <FNF,FSF> pair in a computing system "run" as a concurrent family of processes. Various functions in FNF and FSF are represented by agents that are mutually informable across the <FNF,FSF> pair. The overall functionality of the system depends on the messages passed across from one agent to another.

To each specification defined by SKA there corresponds two modules running concurrent. Since the designs separate out supervision, fault and exception checking onto a concurrent system, the resulting designs tend to be quite efficient from the performance view point. The degree to which the agents are intelligent determines the supervision efficiency and the recovery speed.

IM is a new computing area defined by [Nourani 1997b] with artificial intelligence principles for multimedia. The area for which the paper provides a foundation for is what multimedia computing is bound to be applied at dimensions and computing phenomena unimagined thus far, yet inevitable with the emerging technologies. The principles defined are practical artificial intelligence and its applications to multimedia. Multimedia AI systems are proposed with new computing techniques defined. Multimedia objects and rules and multimedia programming techniques are presented via a new language called IM. Its foundations logic is presented further by [Nourani 1998a]. The basis for the present project is to design multimedia specific chips for computing agents carrying out predefined multimedia tasks. The IM Hybrid Multimedia Programming techniques have a computing logic counterpart called Morph Gentzen since[Nourani 98a] . The basic principles are a mathematical logic where a Gentzen or natural deduction systems is defined by taking multimedia objects coded by diagram functions, foundations published at logic conferences [see Nourani 98a for a technical report]. By trans-morphing hybird picture's corresponding functions a new hybrid picture is deduced. Multimedia objects are viewed as syntactic objects defined by functions, to which the deductive system is applied. Thus we define a syntactic morphing to be a technique by which multimedia objects and hybrid pictures are homomorphically mapped via their defining functions to a new hybrid picture. The deduction rules are a Gentzen system augmented by Morphing, and Trans-morphing. The logical language has function names for hybrid pictures.

The MIM Morph Rule - An object defined by the functional n-tuple <f1,...,fn> can be morphed to an object defined by the functional n-tuple <h(f1),...,h(fn)>, provided h is a homorphism of intelligent objects as abstract algebras[Nourani 93c]. The MIM TransMorph Rules- A set of rules whereby combining hybrid pictures p1,...,pn defines an Event {p1,p2,...,pn} with a consequent hybrid picture p. Thus the combination is an impetus event. The deductive theory is a Gentzen system in which hybrid pictures are named by parameterized functions; augmented by the MIM morph and transmorph rules. The complete formal AI and mathematics is published by Nourani 97, for example[Nournai 98a]. An example trans-morphing with hybird pictures Fi is: <F1> | <F2> & <F3> ===> <F4>.

We have proposed the Morph Gentzen multiagent logic as a basis to design automatic very high speed visual field prediction applied to onboard aerospace and crafts guidance and control. The same tewhcniques are applied to design planetary rovers as proposed by our papers.

We have presented techniques with <object, co-object> pairs and Multiagents cooperating for problem solving on boards as a new AI problem solving paradigm. A model for computer vision is presented based on the present techniques comparable to vision. Intelligent tree computing and linguistic theories from recent research of this author was applied as a start for a theoretical basis for formalizing the computing paradigm.

Further, a basis for a computational epistemics of vision, a new area for research proposed by this author, is presented in brief. This paper is a startup for three areas of research. The three areas consist of the double vision computing paradigm that forms the present extended abstract, the computational epistemics for vision, and the mathematical theory for double vision computing. The techniques for robot supervision design presented has actually been applied to challenging practical problems in system design{Nourani 1995b). Illustrating examples are provided to clarify the design methods. The specific techniques have been a subject of research by the present author over the past few year. There are concept papers by the author referenced which crystallized Fault Tolerant Artificial Intelligence Systems. Faults and exceptional behavior are an essential part of a system and must be thought of at conceptualization [Nourani 1998f]. Such considerations form a basis for robot supervision designs. There are new applications to autonomous spacecraft vision and planetary rovers reported at IV-98 DaimlerBenz, Stuttgart, October 1998.

Fikes, R.E. and N.ilsson, N.J. (1971). Strips: A New Approach to the Application of Theorem Proving to Problem Solving, AI 2, 1971, pp. 189-208.
Genesereth, M. and Nilsson. N.J. (1987). Logical Foundations of Artificial Intelligence, Morgan-Kaufmann, 1987.
Hays-Roth, B (1985). A Balckboard Architecture For Control, Artificial Intelligence, 26(3), 1985, 251-321.
Kirkeboen, G.(1992). Informasjonsteknologien og den cognitive <<revolusjon>> i psykologien, Nordisk AI Magazine, 7, 1992, nr. 2, 4-18.
Nilsson, N.J. (1971). Problem Solving Methods in Artificial Intelligence, 1971, New York, McGraw Hill.
Nourani, C.F. (1984). Equational Intensity, Initial Models, and AI Reasoning, Technical Report, : A Conceptual Overview, in Proc. Sixth European Conference in Artificial Intelligence, Pisa, Italy, September 1984, North- Holland.
Nourani, C.F. (1987). Diagrams, Possible Worlds, and the Problem of Reasoning in Artificial Intelligence, Summer Logic Colloquium, Padova, Italy, 1987.
Nourani, C.F. (1980). A Model-theoretic Approach to Specification, Extension, and Implementation of Data Types, Proc.International Symposium on Programming, Paris, April 1980, Springer-Verlag, Berlin.
Nourani, C.F. (1991). Planning and Plausible Reasoning in AI, Proc. Scandinavian Conference in AI, May 1991, Denmark, 150-157, IOS Press.
Nourani,C.F. (1994). A Theory For Pr ogramming With Intelligent Syntax Trees and Intelligent Decisive Agents, 11th ECAI-94, DAI Applications, " August 1994, Amsterdam.

Nourani, C.F. (1993a). Slalom Tree Computing January 1993 .
AI Communications, Novmebr 1996, IOS Press, Amsterdam

Nourani, C.F. (1993b). A Multiagent Approach To Fault Free and Fault Tolerant AI, Proc. Sixth FLAIRS, Florida AI Research Symposium, April 1993.
Nourani, C.F. (1993c). Abstract Implementations By Computing Agents: A Conceptual Overview," Proc. SERF-93, November 1993, Orlando, FL.

Nourani, C.F. (1993d). Worlds, Models, and Artificial Intelligence Computation,", January 16, 1993.

Nourani,C.F. (1993f). Automatic Models From Syntax,",November 15, 1993, Proc. XV Scandinavian Linguistics Conference, Januray 1995, Oslo, Norway.

Nourani, C.F. (1997) A Computational Epistemics For Computer Vision," Forthcoming.

Nourani, C.F. and Th. Hoppe, (1994). GF Diagrams For Models and Free Proof Trees, Brief Abstract, Presented Berlin Logic Colloquium, Germany, May 1994.

Nourani, C.F.(1994) , "Towards Computational Epistemology- A Forward," Summer Logic Colloquium, Clairmont-Ferrand, France, July 1994.

Nourani,C.F. (1993g). Linguistic Computation, Automatic Model Generation, and Intensions," January 1993; The Computation and Language E-Print Archive; LNL 9409009

Winston, P.H. (1975). The Psychology of Computer Vision, New York, McGraw Hill, 1975, (edited by).1

Nourani,C.F.(1995a),"Double Vision Computing,"IAS-4, Intelligent Autonomous Systems,Karlsruhe, April 1995, Germany

Nourani,C.F.(1995b),"Multiagnet Robot Supervision," Learnign Robots, Herklion, April 1995.

Nourani,C.F.(1995d) Free Porof Trees and Model-Theoretic Planning," AISB, Sheffiled, April 1995.

Nourani, C.F. (19988a) MIM Logik, Summer Logic Colloquium, Prague, August 1998.

Nourani,C.F. (1997b) Nourani, C.F.,"Intelligent Multimedia,"Poster Annoucement Intelligent Multimedia, ICCIMA, Monash University, Gippsland, Vicotoria, Australia, February 1998.

Gini and Gini(1983) Gini, M and G. Gini, "Towards Automatic Recovery in Robot Programs," Proc. 8th IJCAI, Vol.2, 1983, pages 821-823, Karlsruhe, Germany.

Nourani, C.F. (1998f) Desinging Fault Tolerant Systems with Multiagent AI, FTCS98, Munich, Germany, June 1998.