This paper postulates a rigorous method for the construction of classical planning domain models. We describe, with the help of a non-trivial example, a tool supported method for encoding such models. The method results in an `object-centred' specification of the domain that lifts the representation from the level of the literal to the level of the object. Thus, for example, operators are defined in terms of how they change the state of objects, and planning states are defined as amalgams of the objects' states. The method features two classes of tools: for initial capture and validation of the domain model; and for operationalising the domain model (a process we call compilation) for later planning. Here we focus on compilation tools used to generate macros and goal orders to be utilised at plan generation time. We describe them in depth, and evaluate empirically their combined benefits in plan-generation speed-up. The method's main benefit is in helping the modeller to pro...

engineering process in the building of applied AI planning systems. GIPO embodies an object centred approach to planning domain modelling. There are two reasons for providing knowledge engineering support for AI planning: (i) to apply a planning system to a new domain to test the planning system itself (ii) to tackle the end-user problem for the engineer who might be a domain expert but need not necessarily have a specialist knowledge of AI planning. Our research is primarily aimed at developing a method and tools to meet the requirements of the latter case (ii), although the benefits can also be enjoyed by planning experts. 1.

This paper describes a multistrategy learning approach for automatic acquisition of planning operators. The two strategies are: (i) learning operators by observing expert solution traces, and (ii) refining operators through practice in a learning-by-doing paradigm. During observation, OBSERVER uses the knowledge that is naturally observable when experts solve problems, without the need of explicit instruction or interrogation. During practice, OBSERVER 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 learningby -doing operator refinement. Given these inputs, our system automatically acquires the preconditions and effects (including conditional effects and preconditions) of the operators. Our approach has been fully implemented in a system called OBSERVER on top of a non-linear planner PRODIGY.We present empirical results t...

We present Rogue, an integrated planning and executing robotic agent. Rogue is designed to be a roving office gopher, doing tasks such as picking up & delivering mail and returning & picking up library books, in a setup where users can post tasks for the robot to do. We have been working towards the goal of building Rogue as a completely autonomous agent which can learn from its experiences improving its own behaviour. In this paper, we focus on describing Rogue's capabilities in executing and processing perception information, including: (1) the generation and execution of a plan which requires observation to make informed planning decisions, and (2) the monitoring of execution for informed replanning. Rogue is implemented and functional on a real indoor robot.

In this thesis we investigate the relationships between the symbolic and sub-symbolic methods used for controlling agents by artificial intelligence, focusing in particular on methods that learn. In light of the strengths and weaknesses of each approach, we propose a hybridisation of symbolic and subsymbolic methods to capitalise on the best features of each. We implement such a hybrid system, called Rachel which incorporates techniques from Teleo-Reactive Planning, Hierarchical Reinforcement Learning and Inductive Logic Programming. Rachel uses a novel representation of be-haviours, Reinforcement-Learnt Teleo-operators (RL-Tops), which defines the behaviour in terms of its desired consequences but leaves the implementation of the policy to be learnt by reinforcement learning. An RL-Top is an abstract, symbolic description of the purpose of a behaviour, and is used by Rachel both as a planning operator and as the definition of a reward function by which the behaviour can be learnt. Two new

Anyone who has ever driven a car on an icy road is aware of the impact the environment can have on our actions. In order to build effective plans, we must be aware of these environmental conditions and predict the effects they will have on our ability to act. In this paper, we present an application of regression trees that allows a robot to learn action models through experience so that it can make similar predictions. We use this approach to allow a mobile robot to learn models to predict the effects of its navigation actions under various terrain conditions and use them in order to produce efficient plans.

To be e#ective, a navigation planner must have knowledge not only of the e#ects an action will have, but also the e#ects that the environment will have on that action (e.g. the robot may travel more slowly over rough terrain) . To address this issue, we have developed an approach called ERA which uses regression tree induction to learn action models that predict the e#ect terrain conditions will have on a robot's navigation actions. The action models support a high level planner that finds e#cient navigation plans. We present the results of a study which evaluated the performance of ERA in environments with noisy e#ectors and sensors. 1. Introduction An e#ective navigation planner for a mobile robot must be able to deal with large state spaces and take into account the uncertainty in the environment. One common approach to dealing with large state spaces is to use a high level planner that operates on a graph representing an abstract map of the area (Arkin, 1998). Each ...

Knowledge engineering for planning is expensive and the resulting knowledge can be imperfect. To autonomously learn a plan operator definition from environmental feedback, our learning system WISER explores an instantiated literal space using a breadth-first search technique. Each node of the search tree represents a state, a unique subset of the instantiated literal space. A state at the root node is called a seed state. WISER can generate seed states with or without utilizing imperfect expert knowledge. WISER experiments with an operator at each node. The positive state, in which an operator can be successfully executed, constitutes initial preconditions of an operator. We analyze the number of required experiments as a function of the number of the missing preconditions in a seed state. We introduce a naive domain assumption to test only a subset of the exponential state space. Since breadth- rst search is expensive, WISER introduces two search techniques to reorder literals at each l...

Mainstream research in planning assumes that input information is complete and correct. There are branches of research into plan generation with incomplete planning problems and with incomplete domain models. Approaches include gaining knowledge aimed at making the input information complete or building robust planners that can generate plans despite the incompleteness of the input. This paper addresses planning with complete and correct input information, but where the domain models are distributed over multiple agents. The emphasis is on domain model acquisition, i.e. the first approach. The research reported here adopts the view that the agents must share knowledge if planning is to succeed. This implies that a recipient must be able to assimilate the shared knowledge with its own. An algorithm for inducing domain models from example domain states is presented. The paper shows how the algorithm can be applied to knowledge assimilation and discusses the choice of representation for knowledge sharing. The algorithm has been implemented and applied successfully to eight domains. For knowledge assimilation it has been applied to date just to the blocks world.

Mission planning is central to space mission operations and has benefited from advances in model-based planning software, but developing a planning model still remains a difficult task. Mission planning constraints arise from many sources, including simulators and engineering specification documents. Ensuring that these constraints are correctly represented in the planner’s model is a challenge. As mission constraints evolve, planning domain modelers must add and update model constraints efficiently using the available source data, catching errors quickly, and correcting the model. We describe the current state of the practice in designing model-based mission planning tools and the challenges facing model developers. We then propose an Interactive Model Development Environment (IMDE) to configure mission planning systems by integrating modeling and simulation environments to reduce model editing time, generate simulations automatically to evaluate plans, and identify modeling errors automatically by evaluating simulation output.