TacAir-Soar is an intelligent, rule-based system that generates believable "human-like" behavior for large scale, distributed military simulations. The innovation of the application is primarily a matter of scale and integration. The system is capable of executing most of the airborne missions that the United States military flies in fixed-wing aircraft. It accomplishes this by integrating a wide variety of intelligent capabilities, including real-time hierarchical execution of complex goals and plans, communication and coordination with humans and simulated entities, maintenance of situational awareness, and the ability to accept and respond to new orders while in flight. The system is currently deployed at the Oceana Naval Air Station WISSARD Lab and the Air Force Research Laboratory in Mesa, AZ. Its most dramatic use was in the Synthetic Theater of War 1997, which was an operational training exercise that ran for 48 continuous hours during which TacAir-Soar flew all U.S. fixed-wing aircraft.

This paper reports on experiments where techniques of supervised machine learning are applied to the problem of planning. The input to the learning algorithm is composed of a description of a planning domain, planning problems in this domain, and solutions for them. The output is an efficient algorithm --- a strategy --- for solving problems in that domain. We test the strategy on an independent set of planning problems from the same domain, so that success is measured by its ability to solve complete problems. A system, L2Act, has been developed in order to perform these experiments. We have experimented with the blocks world domain, and the logistics domain, using strategies in the form of a generalization of decision lists, where the rules on the list are existentially quantified first order expressions. The learning algorithm is a variant of Rivest`s [39] algorithm, improved with several techniques that reduce its time complexity. As the experiments demonstrate, generalization is a...

Intelligent artificial agents need to be able to explain and justify their actions. They must therefore understand the rationales for their own actions. This paper describes a technique for acquiring this understanding, implemented in a multimedia explanation system. The system determines the motivation for a decision by recalling the situation in which the decision was made, and replaying the decision under variants of the original situation. Through experimentation the agent is able to discover what factors led to the decisions, and what alternatives might have been chosen had the situation been slightly different. The agent learns to recognize similar situations where the same decision would be made for the same reasons. This approach is implemented in an artificial fighter pilot that can explain the motivations for its actions, situation assessments, and beliefs. Introduction Intelligent artificial agents need to be able to provide explanations and justifications for the actions t...

The origins of the Soar architecture can be traced back to the seminal research of Allen Newell and Herbert Simon on symbol systems, heuristic search, goals, problem spaces, and production systems. Since its official inception in 1982, Soar has evolved through six major releases, as both an AI architecture and as the basis for a unified theory of cognition. This paper traces this evolutionary path, starting with Soar's intellectual roots, and then proceeding through the stages defined by the six major system releases. Each stage is characterized with respect to a hierarchy of four levels of analysis: the knowledge level, the problem space level, the symbolic architecture level, and the implementation level.

This paper describes an organizing conceptual structure for current real-time AI research, clarifying the different meanings this term has acquired for various researchers

The utility problem in learning systems occurs when knowledge learned in an attempt to improve a system's performance degrades performance instead. We present a methodology for the analysis of utility problems which uses computational models of problem solving systems to isolate the root causes of a utility problem, to detect the threshold conditions under which the problem will arise, and to design strategies to eliminate it. We present models of case-based reasoning and control-rule learning systems and compare their performance with respect to the swamping utility problem. Our analysis suggests that case-based reasoning systems are more resistant to the utility problem than control-rule learning systems. 1 1. Introduction An interesting asymmetry exists in the patterns of retrieval in case-based reasoning (CBR) and control-rule learning (CRL) systems: to take advantage of past learning experiences, CRL systems need to retrieve rules from memory at each step, whereas CBR systems ne...

Abstract not found

This paper describes CES, a proto-type of a new programming language for robots and other embedded systems, equipped with sensors and actuators. CES contains two new ideas, currently not found in other programming languages: support of computing with uncertain information, and support of adaptation and teaching as a means of programming. These innovations facilitate the rapid development of software for embedded systems, as demonstrated by a mobile robot application.

In complex, dynamic environments, an agent's domain knowledge will rarely be complete and correct. Existing deliberate approaches to domain theory correction are significantly restricted in the environments where they can be used. These systems are typically not used in agent-based tasks and rely on declarative representations to support non-incremental learning. This research investigates the use of procedural knowledge to support deliberate incremental error correction in complex environments. We describe a series of domain properties that constrain the error correction process and that are violated by existing approaches. We then present a procedural representation for domain knowledge which is sufficiently expressive, yet tractable. We develop a general framework for error detection and correction and then describe an error correction system, IMPROV, that uses our procedural representation to meet the constraints imposed by complex environments. Finally, we test the syst...

In systems which learn a large number of rules (productions), it is important to match the rules efficiently, in order to avoid the machine learning utility problem --- if the learned rules slow down the matcher, the "learning" can slow the whole system down to a crawl. So we need match algorithms that scale well with the number of productions in the system. (Doorenbos, 1993) introduced right unlinking as a way to improve the scalability of the Rete match algorithm. In this paper we build on this idea, introducing a symmetric optimization, left unlinking, and demonstrating that it makes Rete scale well on an even larger class of systems. Unfortunately, when left and right unlinking are combined in the same system, they can interfere with each other. We give a particular way to combine them which we prove minimizes this interference, and analyze the worst-case remaining interference. Finally, we present empirical results showing that the interference is very small in practice, and that...