The paper proposes that all information systems can be treated as thermodynamic engines, therefore their performance can be measured, evaluated and compared with observable thermodynamically-related properties. This proposition suggests a simple, parsimonious method for characterizing information systems using physically measurable properties to represent their performance. Measurements of these properties can compare, both absolutely and relatively, the performance of information systems performing identical tasks, independent of their designs. These measurements are realistically accessible, come at reasonable cost, and can be performed in an acceptable amount of time. This method derives directly from the physical dependencies between the energy based and information based worlds and explores the performance of entire systems as well as their individual components.

A recent NATO advanced research workshop on mobile robot implementation is reported. This workshop was focused on the problems inherent to mobile robot implementation which have not been widely discussed in the research community. Several important problem areas were defined by these discussions including sensors, communications, vehicle modeling, software development tools, human interfaces for teleoperated vehicles, techniques for performance evaluation and configuration management. These problems may seriously limit the applications for which mobile robots can be practically implemented.>

A number of economoic and social forces are driving the metamorphosis of today's automated manufacturing technology into a future autonomous manufacturing capability. However, advancements in both offline planning and real time operations are necessary before actual autonomous facilities are realized. Much work is currently addressing all of these technical areas in the United States but considerable distance still separates today's flexible manufacturing implementations from tomorrow's autonomous manufacturing capabilities.

The Ground Surveillance Robot (GSR) project has proceeded continuously since the Fall of 1980, and in that time an autonomous vehicle design and some degree of implementation has been achieved. The vehicle design has been partitioned into sensor, control, and planning subsystems. A distributed blackboard scheme has been developed which provides the mechanism by which these subsystems are coordinated. Vehicle position and orientation are supplied by vehicle attitude and navigation sensor subsystems. Obstacle avoidance capability has been implemented by fusing information from vision and acoustic ranging sensors into local goals and avoidance points. The influence of these points is combined through potential field techniques to accomplish obstacle avoidance control. Distant terrain characteristics are identified using information from a gray-level vision system, a color vision system, and a computer-controlled laser ranging sensor. These characteristics are used by a general planning engine to develop the desired path to a visible goal in the direction of the final goal. Progress to the final goal consists of a succession of movements from one distant but visible intermediate goal to another. The experience from implementing this autonomous vehicle has indicated the need for an integrated set of debugging tools which make the faults in subsystem hardware and software more distinguishable.

This paper describes a technique for coordinating the subsystems of autonomous robots which takes advantage of a distributed blackboard mechanism and a high degree of functional distribution between subsystems to minimize communications and simplify the interfaces. Distributed blackboard memory contains a world model which represents knowledge about itself and its surroundings as collections of objects important to the task and the relations between them. Objects or instances are represented as lists of object-attribute-value-accuracy-confidence-timestamp tuples which are organized into a class tree with inheritance properties and active functions. Intelligent Communications Interfaces maintain the consistency between blackboard memory elements distributed through loosely coupled subsystems by exchanging reports which represent world state information and plans which represent control information. These concepts can be extended for coordinating multiple autonomous robots by making independent robot world models intersect to enable communications and by broadening the interpretation of plans and reports. Coordinating autonomous robots through distributed blackboards makes potentially complex programming simpler by providing a well defined but flexible framework for module interfaces.

Sensor data fusion takes many forms and has diverse purposes. Sensor data fusion can improve a complex robot's picture of the world and increase its confidence in the truth of that picture. A computing architecture which supports many different schemes for fusing sensor data is necessary to support the activities of complex robots. One such architecture is the distributed blackboard mechanism implemented onboard the USMC Ground Surveillance Robot (GSR) [1]. The distributed blackboard has proven to be a very useful and flexible mechanism through which to accomplish effective sensor data fusion.

This brief paper provides one person's perspective on route planning problems for autonomous mobile robots operating in unknown natural terrains. These problems include representation of geometric and nongeometric conditions, reasoning in dynamic situations, obstacle sensing problems and incomplete world knowledge. Several researchers' attacks on these problems are reviewed.