are those of the authors and should not be interpreted as necessarily representing the official policies or endorsements,

Abstract not found

An important area of learning in autonomous agents is the ability to learn domain-specific models of actions to be used by planning systems. In this paper, we present methods by which an agent learns action models from its own experience and from its observation of a domain expert. These methods differ from previous work in the area in two ways: the use of an action model formalism which is better suited to the needs of a reactive agent, and successful implementation of noise-handling mechanisms. Training instances are generated from experience and observation, and a variant of GOLEM is used to learn action models from these instances. The integrated learning system has been experimentally validated in simulated construction and office domains. 1 INTRODUCTION Autonomous agents acting in complex environments must be capable of learning from experience, both to avoid the need for exhaustive preprogramming and to adapt to unanticipated or changing situations. Most such work has focused o...

Acquiring and maintaining domain knowledge is a key bottleneck in applications of planning systems. This thesis describes a machine learning approach to automatic acquisition of planning operators. Our approach is to learn planning operators by observing expert solution traces and to refine operators through practice in a learning-by-doing paradigm. During observation, our system uses the knowledge that is observable when experts solve problems, without the need of explicit instruction or interrogation. During practice, our system generates its own learning opportunities by solving practice problems. The inputs to our learning system are: the description language for the domain, experts' problem solving traces, and practice problems to allow learning-by-doing operator refinement. The output is a set of operators, each described by a list of variables, preconditions, and effects. The operators are learned incrementally using an inductive algorithm. During practice, our system effective...

We present an autonomous agent architecture and its component subsystems that integrate important abilities needed for robust, flexible performance in dynamic environments. These abilities involve appropriate reaction to environmental situations given the agent's goals; selective attention to multiple, competing goals; planning new action routines when innovation beyond designer-provided routines is necessary; and learning the effects of actions so that the planner can use them to build ever more reliable plans. The teleo-reactive format allows actions to be closely coupled to continuous environmental feedback and is also especially compatible with conventional AI planning and learning mechanisms. The workings of the proposed architecture and its subsystems are illustrated in a simulated robot domain. We conclude by noting areas where future work is needed.

Generating production-quality plans is an essential element in transforming planners from research tools into real-world applications. However most of the work to date on learning planning control knowledgehas beenaimed at improving the efficiency of planning; this work has been termed "speed-up learning". This paper focuses on learning control knowledge to guide a planner towards better solutions, i.e. to improve the quality of the plans produced by the planner, as its problem solving experience increases. We motivate the use of quality-enhancing search control knowledge and its automated acquisition from problem solving experience. We introduce an implemented mechanism for learning such control knowledge and some of our preliminary results in a process planning domain. Introduction Most research on planning so far has concentrated on methods for constructing sound and complete planners that find a satisficing solution, andonhow to find such solution in an efficient way (Chapman 198...

Thispaper describesissues that arise when integrating a planner with a system that learns planning operators incrementally, and our approaches to address these issues. During learning, domain knowledge can be incomplete and incorrect in different ways; therefore the planner must be able to use incomplete domain knowledge. This presents the following challenges for planning: How should the planner effectively generate plans using incompleteand incorrect domain knowledge? How should the planner repair plans upon execution failures ? How should planning, learning, and execution be integrated? This paper describes how we address these challenges in the framework of an integrated system, called OBSERVER, that learns planning operators automatically and incrementally. In OBSERVER, operators are learned by observing expert agents and by practicing in a learning-by-doing paradigm. We present empirical results to demonstrate the validity of our approach in a process planning domain. These resu...

Many advanced safety and navigation applications in vehicles require accurate, detailed digital maps, but manual lane measurements are expensive and time-consuming, making automated techniques desirable. This paper describes a data-mining approach to map refinement, using position traces that come from Global Positioning System receivers with differential corrections. The computed lane models enable safety applications, such as lanekeeping, and convenience applications, such as lane-changing advice. Experiments show that, starting from a baseline map that is commercially available, our lane models predict a vehicle's lane with high accuracy from a small number of passes over a particular road segment. Multiple position traces are a powerful new source of data that enables cheap, automated methods of inducing lane models, as well as other geographic knowledge, like traffic signals and elevations, and potentially impacts any geographic information system with a need to relate to actual b...

A great challenge in using any planning system to solve real-world problems is the difficulty of acquiring the domain knowledge that the system will need. We present a way to address part of this problem, in the context of Hierarchical Task Network (HTN) planning, by having the planning system incrementally learn conditions for HTN methods under expert supervision. We present a general formal framework for learning HTN methods, and a supervised learning algorithm, named CaMeL, based on this formalism. We present theoretical results about CaMeL’s soundness, completeness, and convergence properties. We also report experimental results about its speed of convergence under different conditions. The experimental results suggest that CaMeL has the potential to be useful in real-world applications.

While recent decades have brought substantial change to the form of the operating system interface, the power of operating system commands has remained nearly constant. Conventional commands, whether visual or textual, specify one particular action to perform. To carry out a complex task, such as reducing disk utilization, the user is forced to explicitly specify each of the necessary steps. Traditional command-language extension mechanisms, such as shell scripts in Unix, enable the user to aggregate and compose various commands, but force him or her to write and debug programs --- a formidable challenge for naive users. This paper presents a goal-oriented approach to the operating system command interface, realized through an implementation we call OS agents. Using OS agents, the user simply specifies a goal to accomplish, and the OS agent decides how to accomplish that goal using its knowledge base of the system state and its commands. The OS agent dynamically synthesizes the appropr...