Fields 3-8 of table 1 of the survey and general results, a discussion represent purposes, specifically, to define of the four hypotheses, and two sections models (field 3), prove theorems about the at the end of the article that contain details of models (field 4), present algorithms (field 5), the survey and statistical analyses. The next analyze algorithms (field 6), present systems section (The Survey) briefly describes the 16 or architectures (field 7), and analyze them substantive questions I asked about each (field 8). These purposes are not mutually paper. One of the closing sections (An Explanation exclusive; for example, many papers that of the Fields in Table 1) discusses the present models also prove theorems about criteria for answering the survey questions the models.

Embedded agents are computer systems that sense and act on their environments, monitoring complex dynamic conditions and affecting the environment in goal-directed ways. This paper briefly reviews the situated automata approach to agent design and explores issues of planning and action in the situated-automata framework.

"We describe a robot control architecture which combines a stimulus-response subsystem for rapid reaction, with a search-based planner for handling unanticipated situations. The robot agent continually chooses which action it is to perform, using the stimulusresponse subsystem when possible, and falling back on the planning subsystem when necessary. Whenever it is forced to plan, it applies an explanation-based learning mechanism to formulate a new stimulus-response rule to cover this new situation and others similar to it. With experience, the agent becomes increasingly reactive as its learning component acquires new stimulus-response rules that eliminate the need for planning in similar subsequent situations. This Theo-Agent architecture is described, and results are presented demonstrating its ability to reduce routine reaction time for a simple mobile robot from minutes to under a second."In AAAI-90, Vol. 2, pp. 1051– 1058

Universal plans address the tension between reasoned behavior and timely response by caching reactions for classes of possible situations. This technique reduces the average time required to select a response at the expense of the space required to store the cache-the classic time-space trade-off. In his article, Matthew Ginsberg argues from the time extreme and against the space extreme. Although I find both extremes undesirable, I defend an increase in space consumption.

Phoenix is a real-time, adaptive planner that manages forest fires in a simulated environment. Alternatively, Phoenix is a search for functional relationships between the designs of agents, their behaviors, and the environments in which they work. In fact, both characterizations are appropriate and together exemplify a research methodology that emphasizes complex, dynamic environments and complete, autonomous agents. This article describes the underlying methodology and illustrates the architecture and behavior of Phoenix agents.

Second, These sections describe how Phoenix agents there are motivating issues, of plan in real time but do not provide the which the foremost is to understand minute detail that is offered elsewhere (Cohen how complex environments et al. forthcoming). The next section illustrates constrain on the design of Phoenix agents controlling a forest fire. We seek general The last section describes the current status of rules that justify and explain the project and our immediate goals. The terms in these rules describe The Phoenix task is to control simulated characteristics of environments, forest fires by deploying simulated bulldozers, tasks and behaviors, and the crews, airplanes, and other objects. We discuss architectures of agents. Phoenix Environment, Layers 1 and 2 but Phoenix is a commentary on the Phoenix Simulator. In the following pages, we describe Phoenix from the perspective of our technical aims and motives. The second section describes the Phoenix task--controlling simulated forest fires-- and explains why we use a simulated environment instead of a real, physical one. The two lowest layers of Phoenix, described in The Phoenix Environment, Layers 1 and 2, implement the simulated environment and maintain the illusion that the forest fire and agents are acting simultaneously. Above these layers are two others: a Figure 2. Fire at 12:30 Bulldozers are Close to organization of multiple Meeting at the Fire Front. The left pane displays the real world; the right pane displays fireboss sees it. Firefighting objects are also and other agents are semiautonomous.

To demonstrate these methods we have trained an ANS network to drive a vehicle through simulated rreeway traffic. I ntJooducticn Computational systems employing fine grained parallelism are revolutionizing the way we approach a number or long standing problems involving pattern recognition and cognitive processing. The field spans a wide variety or computational networks, rrom constructs emulating neural runctions, to more crystalline configurations that resemble systolic arrays. Several titles are used to describe this broad area or research, we use the term artificial neural systems (ANS). Our concern in this work is the use or ANS ror manually training certain types or autonomous systems where the desired rules of behavior are difficult to rormulate. Artificial neural systems consist of a number or processing elements interconnected in a weighted, user-specified fashion, the interconnection weights acting as memory ror the system. Each processing element calculatE', an output value based on the weighted sum or its inputs. In addition, the input data is correlated with the output or desired output (specified by an instructive agent) in a training rule that is used to adjust the interconnection weights.

To demonstrate these methods we have trained an ANS network to drive a vehicle through simulated rreeway traffic. I ntJooducticn Computational systems employing fine grained parallelism are revolutionizing the way we approach a number or long standing problems involving pattern recognition and cognitive processing. The field spans a wide variety or computational networks, rrom constructs emulating neural runctions, to more crystalline configurations that resemble systolic arrays. Several titles are used to describe this broad area or research, we use the term artificial neural systems (ANS). Our concern in this work is the use or ANS ror manually training certain types or autonomous systems where the desired rules of behavior are difficult to rormulate. Artificial neural systems consist of a number or processing elements interconnected in a weighted, user-specified fashion, the interconnection weights acting as memory ror the system. Each processing element calculatE', an output value based on the weighted sum or its inputs. In addition, the input data is correlated with the output or desired output (specified by an instructive agent) in a training rule that is used to adjust the interconnection weights.

To demonstrate these methods we have trained an ANS network to drive a vehicle through simulated rreeway traffic. I ntJooducticn Computational systems employing fine grained parallelism are revolutionizing the way we approach a number or long standing problems involving pattern recognition and cognitive processing. Thefield spans a wide variety or computational networks, rrom constructs emulating neural runctions, to more crystalline configurations that resemble systolic arrays. Several titles are used to describe this broad area or research, we use the term artificial neural systems (ANS). Our concern inthis work is the use or ANS ror manually training certain types or autonomous systems where the desired rules of behavior are difficult to rormulate. Artificial neural systems consist of a number or processing elements interconnected in a weighted, user-specified fashion, the interconnection weights acting as memory ror the system. Each processing element calculatE', an output value based on the weighted sum or its inputs. In addition, the input data is correlated with the output or desired output (specified by an instructive agent) in a training rule that is used to adjust the interconnection weights.

The Robotics Institute, Carnegie Mellon University Pittsburgh, PA 15213 Abstract In non-cooperative multi-agent planning, resolution of multiple conflicting goals is the result of finding compromise solutions. Previous research has dealt with such multi-agent problems where planning goals are well-specified, subgoals can be enumerated, and the utilities associated with subgoals known. Our research extends the domain of problems to include non-cooperative multi-agent interactions where planning goals are ill-specified, subgoals cannot be enumerated, and the associated utilities are not precisely known. Negotiation is performed through proposal and modification of goal relaxations. Case-Based Reasoning is integrated with the use of multi-attribute utilities to portray tradeoffs and propose novel goal relaxations and compromises. Persuasive arguments are generated and used as a mechanism to dynamically change the agents' utilities so that convergence to an acceptable compromise can be achieved.