PRISM :: Browsing by Author "MacDonald, Bruce A."

Browsing by Author "MacDonald, Bruce A."

Now showing 1 - 20 of 25

Open Access
A computational model of task acquisition from instruction
(1992) Lewis, John A.; MacDonald, Bruce A.
Open Access
Adaptive robot training: explorations in sensorless manipulation
(1990) Pauli, David; MacDonald, Bruce A.
Open Access
AUTONOMY, INTELLIGENCE, AND INSTRUCTABILITY
(1988-10-01) Witten, Ian H.; MacDonald, Bruce A.
Instructable systems constitute an important, useful, and practically realizable step towards fully autonomous ones. In many applications people will not want machines to be self-motivated, but they will want to teach them new jobs. The user interface must permit the teacher to guide the system through tasks. The system employs samples of behavior so gathered to drive an inductive process of concept learning. Learning becomes intractable unless the teacher fulfils certain felicity conditions. The real world frequently constitutes a competitive learning environment, and instructable systems may have to guard against their knowledge and skills being corrupted by incorrect or deliberately misleading teachers. Experimental prototypes of two instructable systems are presented, one for verbally editing robot movements, the other for automating office tasks. These examples show the potential utility of approaching autonomy via instructability; the next steps are to extend the power of their learning mechanisms, and to render them robust.
Open Access
CONCEPT LEARNING: A PRACTICAL TOOL FOR KNOWLEDGE ACQUISITION?
(1987-01-01) Witten, Ian H.; MacDonald, Bruce A.
Knowledge acquisition has emerged as a key technology underpinning the creation of expert systems. A number of schemes have appeared recently which learn structural concepts from examples, thereby offering the potential to interact directly with domain experts to acquire knowledge. This paper examines the fundamental distinctions which underlie current research, relates them to plain questions about the capabilities of learning systems, and critically reviews representative systems from the perspective of practical knowledge acquisition.
Open Access
CONNECTING TO THE PAST
(1988-01-01) MacDonald, Bruce A.
Recently there has been renewed interest in neural-like processing systems, evidenced for example in the two volumes Parallel Distributed Processing edited by Rumelhart and McClelland, and discussed as parallel distributed systems, connectionist models, neural nets, value passing systems and multiple context systems. Dissatisfaction with symbolic manipulation paradigms for artificial intelligence seems partly responsible for this attention, encouraged by the promise of massively parallel systems implemented in hardware. This paper relates simple neural-like systems based on multiple context to some other well-known formalisms--namely production systems, k-length sequence prediction, finite-state machines and Turing machines--and presents earlier sequence prediction results in a new light.
Open Access
Demonstration instead of programming: focussing attention in robot task acquisition
(1989) Heise, Rosanna; MacDonald, Bruce A.
Open Access
DoLittle: a learning multi-strategy planner
(1996) Baltes, Hansjorg; MacDonald, Bruce A.
Open Access
DYNAMIC BIAS IS NECESSARY IN REAL WORLD
(1991-04-01) Heise, Rosanna; MacDonald, Bruce A.
This paper discusses the bias present in machine learning systems, emphasizing its effect on learnability and complexity. A good bias must allow more concepts to be learned and/or decrease the complexity associated with learning. The paper develops an exhaustive framework for bias, with two important distinctions: \fIstatic\fR versus \fIdynamic\fR and \fIfocus\fR versus \fImagnify\fR. The well-known candidate elimination algorithm (Mitchell) is used to illustrate the framework. Real world learners need dynamic bias. The paper examines two representative systems. \s+2S\s-2TABB (Utgoff) dynamically magnifies the description space where learning would otherwise be impossible. \s+2E\s-2TAR is a prototype for learning robot assembly tasks from examples--a dynamic focusing mechanism reduces both the real world description space and the task construction complexity. Inductive learning must be viewed as a problem of dynamic search control.
Open Access
A FRAMEWORK FOR KNOWLEDGE ACQUISITION THROUGH TECHNIQUES OF CONCEPT LEARNING
(1988-03-01) MacDonald, Bruce A.; Witten, Ian H.
Knowledge-based systems must represent information abstractly so that it can be stored and manipulated effectively. Schemes for learning suitable representations--or concepts--from examples promise domain experts direct interaction with machines to transfer their knowledge. This paper develops an integrative framework for describing concept learning techniques which enables their relevance to knowledge engineering to be evaluated. The framework provides a general basis for relating concept learning to knowledge acquisition, and is a starting point for the development of formal design rules. The paper first frames concept learning in the context of knowledge acquisition. It then discusses the general forms of input and concept representation: as logic, functions and procedures. Next, methods of biasing the search for a suitable concept are described and illustrated: background knowledge, conceptual bias, composition bias, and preference orderings. Then modes of teacher interaction are reviewed: the nature of examples given, and the method of presenting them. Finally the framework is illustrated by applying it to the better-documented concept learning systems.
Open Access
Height simulation of dynamically balanced bipeds
(1991) Williams, Mark Stephen; MacDonald, Bruce A.
This thesis investigates a series of models of a dynamically balanced bipedal running robot. The simulation models are used to study the control of running height to aid the building of dynamically balanced legged systems. A simple mass on a spring model evolved into a complex system based on the use of a PID (ProportionalÂIntegral-Derivative) controlled hydraulic actuator to inject energy into an air spring. The behaviour of hydraulic systems are nonlinear requiring detailed simulation. The simulation of these models is based on the study of the free body forces acting on the system. From the sum of these forces, a net acceleration is determined. A general simulation environment based on a fourth-order Runge-Kutta method, was implemented to solve the differential equations of motion. The thesis resulted in a biped model that can be accurately controlled to achieve a running height based only on the specification of a desired height.
Open Access
INSTRUCTABLE SYSTEMS
(1990-10-01) MacDonald, Bruce A.
Instructable systems - both instructable robots and instructable agents - must acquire skills and knowledge from examples and other instructions easily given by users in factories, laboratories and offices. Both senses of "instruct" are important: command and teach. The human interface must exploit the user's natural instruction abilities and require minimal acquisition of expertise prior to teaching. It is assumed that typical users will not be expert programmers, but will be able to do the tasks they wish to teach and also show them to other humans. Inductive learning techniques are employed to generalize the teacher's examples, in a manner biased by the teacher's other instructions, and thereby form a procedural task description. Instructions can drastically reduce the example and computational complexities of learning problems without compromising learnability. Existing machine learning systems are placed in an instructable framework. Three experimental prototypes are briefly described. Two systems instruct robots: one emphasizing examples and the other emphasizing more explicit instructions. The third is an instructable, office clerk metaphor. Instructability is seen as a small, but significant step toward intelligence.
Open Access
KINEMATICS OF AN ELBOW MANIPULATOR WITH FOREARM ROTATION: THE EXCALIBUR
(1988-10-01) Heise, Rosanna; MacDonald, Bruce A.
General methods and typical specific solutions for robot arm geometry are well-known. This paper presents a detailed solution for a six joint manipulator which has a rotation at mid-forearm rather than a third wrist axis. Details are given indicating how real joint angles relate to those modeled by the more abstract kinematics. All degeneracies are considered and methods for handling them are given. The paper provides a complete tutorial for kinematic modeling with a specific arm.
Open Access
L-EBE, learning iterative editing by example
(1993) Schuler, Natascha Olivia; MacDonald, Bruce A.
Open Access
Optimal tunneling: a heuristic for learning macro operators
(1993) James, Mark L.; MacDonald, Bruce A.
Open Access
THE OTHER SIDE OF THE COIN: TEACHING ARTIFICIAL LEARNING SYSTEMS
(1988-09-01) Witten, Ian H.; MacDonald, Bruce A.
The burgeoning technology of machine learning is beginning to provide some insight into the nature of learning and the role of teaching in expediting the learning process. A number of systems that learn concepts and procedures from examples have been described in the research literature. In general these require a teacher who not only has an analytical understanding of the problem domain, but also is familiar with some of the internal workings of the learning system itself. This is because the learner is performing a search in concept space which is generally quite intractable, but for the teacher's selection of guiding examples. A concept learning system's teacher must select a complete, properly ordered set of examples-one that results in a successful search by the system for an appropriate concept description. In some systems the set of examples determines whether the concept can or cannot be learned, while the order of presentation affects execution time alone. In others, both examples and presentation order are jointly responsible for success. Yet others occasionally select critical examples themselves and present them to a teacher for classification. In all cases, however, the teacher provides the primary means whereby search is pruned. Sometimes the teacher must prime the learner with considerable initial knowledge before learning can begin. Not surprisingly, systems which demand more of the teacher are able to learn more sophisticated concepts. This paper examines the relationship between teaching requirements and learning power for current concept learning systems. We introduce concept learning by machine with emphasis on the role of the human teacher in rendering practical an otherwise intractable concept search. Machine learning has drawn many lessons from human learning and will continue to do so. In turn it can contribute more formal, if simpler, analysis of concept learning from examples.
Open Access
PROGRAMMING BY EXAMPLE: THE HUMAN FACE OF AI
(1991-04-01) Witten, Ian H.; MacDonald, Bruce A.; Maulsby, David L.; Heise, Rosanna
Interactive computer users often find themselves repeatedly performing similar tasks that could be acquired automatically from a teacher. This paper presents principles derived from experience in creating four prototype learners: for technical drawing, text editing, office tasks, and robot assembly. A teaching metaphor (a) enables the user to demonstrate a task by performing it manually, (b) helps to explain the learner's limited capabilities in terms of a persona, and (c) allows users to attribute intentionality. Tasks are represented procedurally, and augmented with constraints. Suitable mechanisms for attention focusing are necessary in order to control inductive search. Hidden features of a task should be made explicit so that the learner need not entertain, and search, all possible missing steps. Key features of the interaction are formalized as "felicity conditions" that help a learner by guaranteeing more explicit, consistent information in demonstrations. Systems that are programmed by human instruction can capitalize on appropriate interactive methods to boost the computational limitations of inductive inference.
Open Access
PROGRAMMING COMPUTER CONTROLLED SYSTEMS BY NON-EXPERTS
(1987-07-01) MacDonald, Bruce A.; Witten, Ian H.
Existing computer-controlled systems lack both teachability and adaptability. It is difficult for end users to specify new procedures to them, or even to modify old ones. This paper introduces two case studies of how non-programmers may define and modify procedures, an activity they are encouraged to conceptualize as teaching rather than programming. In contrast to many AI approaches to learning, the emphasis is on making the systems accessible through a suitably-engineered human-machine interface. The first system allows a teacher to correct a previously-taught robot action sequence by making on-line adjustments using his natural verbal ability. The second is a programming metaphor for office users which complements the object-oriented method of conventional direct manipulation systems by providing an "office clerk" that the user can instruct by giving examples and commands.
Open Access
PROGRAMMING COMPUTER CONTROLLERS BY GIVING EXAMPLES
(1987-06-01) MacDonald, Bruce A.
It is difficult for non-experts to specify a new procedure to a computer controlled system. The problem is that non-programmers, and sometimes skilled programmers, have difficulty in setting up new tasks. If we expect robots and other programmable computer controlled systems to expand in the market place, then the set up cost for new tasks must be reduced. Specifying new procedures to a system controller is difficult because existing systems need a description \fIin a programming language.\fR People do not find it easy to translate their natural knowledge of a task into programs. Some existing robots can be led through a fixed sequence of movements, which are recorded for later repetition (e.g. spray-painting robots). Thus a user can specify that a fixed sequence be automated, without needing programming expertise. However, once the sequence must be edited, the user must employ some programming expertise. This paper examines a technique for editing robot sequences verbally. This verbal correction scheme would enable a teacher to make online adjustments to a robot's action sequence, using his natural verbal ability. As well as editing, conditionals, iteration and data structures are required in procedures. Methods for a user to naturally specify these are briefly discussed.
Open Access
QUATERNIONS AND MOTION INTERPOLATION
(1988-11-01) Heise, Rosanna; MacDonald, Bruce A.
This paper explains straight-line interpolation of solid object motion, such as robot end effector translation and rotation. Smoothly changing orientation is accomplished using quaternions- a way of representing every orientation as four numbers (an angle and an axis of rotation). The first portion of the paper clarifies quaternions to provide an intuitive understanding of their role in rotation. Interpolation is then discussed, concluding with some problems in real manipulator implementations. The interpolation method has been tested on an Excalibur robot.
Open Access
ROBOTS ACQUIRING TASKS FROM EXAMPLES
(1988-12-25) MacDonald, Bruce A.; Heise, Rosanna
This paper describes a task acquisition system which is being implemented on a six-joint robot. Functions controlling the robot are constructed directly from examples of the user leading it. The numerical robot feedback is passed through a symbolic processing stage to convert it into primitive motion functions. Thereafter, generalization occurs at two levels - the primitive motion function names and the arguments to these primitive functions. The constructed task function may contain loops, conditionals, and variables. All variables are determined from the objects which are manipulated. General algorithms are described, examples are given, and comparisons to existing operator learning systems are presented.