Despite the extensive development of first-principles planning in recent years, planning applications are still primarily developed using knowledge-based planners which can exploit domain-specific heuristics and weaker domain models.  Hierarchical Task Network (HTN) planners capture domain-specific heuristics for more efficient search, accommodate incomplete causal models, and can be used to enforce standard operating procedures.  Unfortunately, we do not have semantics for the methods or tasks that make up HTN models, that help evaluate the correctness of methods, or to build a reliable executive for HTN plans.  This paper fills the gap by providing a well-defined semantics for the methods and plans of SHOP2, a state-of-the-art HTN planner.  The semantics are defined in terms of concurrent golog (ConGolog) and the situation calculus.  We provide a proof of equivalence between the plans generated by SHOP2 and the action sequences of the ConGolog semantics.  We show how the semantics reflects the distinction between plan-time and execution-time, and provide some simple examples showing how the semantics can support method verification.  The semantics provide an implementation-neutral specification for an executive, showing how an executive must treat the plans SHOP2 generates in order to enforce the expected behaviors.  Future directions include automated verification of method specifications, automatically generating plan monitors, and plan revision and repair.

The problem of formulating knowledge bases containing action schema is a central concern in knowledge engineering for AI Planning. This paper describes LOCM, a system which carries out the automated induction of action schema from sets of example plans.  Each plan is assumed to be a sound sequence of actions; each action in a plan is stated as a name and a list of objects that the action refers to. LOCM exploits the assumption that actions change the state of objects, and require objects to be in a certain state before they can be executed.  The novelty of LOCM is that it can induce action schema without being provided with any information about predicates or initial, goal or intermediate state descriptions for the example action sequences.  In this paper we describe the implemented LOCM algorithm, and analyse its performance by its application to the induction of domain models for several domains. To evaluate the algorithm, we used random action sequences from existing models of domains, as well as solutions to past IPC problems.

In this paper, we address the problem of generating preferred plans by combining the procedural control knowledge specified by Hierarchical Task Networks (HTNs) with rich qualitative user preferences. The outcome of our work is a language for specifyin user preferences, tailored to HTN planning, together with a provably optimal preference-based planner, HTNPLAN, that is implemented as an extension of SHOP2. To compute preferred plans, we propose an approach based on forward-chaining heuristic search. Our heuristic uses an admissible evaluation function measuring the satisfaction of preferences over partial plans. Our empirical evaluation demonstrates the effectiveness of our HTNPLAN heuristics. We prove our approach sound and optimal with respect to the plans it generates by appealing to a situation calculus semantics of our preference language and of HTN planning. While our implementation builds on SHOP2, the language and techniques proposed here are relevant to a broad range of HTN planners.

Constraint Programming is an attractive approach for solving AI planning problems by modelling them as Constraint Satisfaction Problems (CSPs). However, formulating effective constraint models of complex planning problems is challenging,  and CSPs resulting from standard approaches often require further enhancement to perform well. Common subexpression elimination is a computationally cheap and  general technique for improving CSPs, which can lead to a great reduction in instance size, solving time and search space. In this work we identify general causes of common subexpressions from three modelling techniques often used to encode planning problems into   constraints. We present four case studies of constraint  models of AI planning problems. In each, we describe the constraint model, highlight the sources of common subexpressions, and present an empirical analysis of the effects of eliminating common subexpressions.

State abstraction has been widely used for state aggregation in approaches to AI search and planning. In this paper we use a powerful abstraction technique from software model checking for representing collections of states with different object quantities and properties. We exploit this method to develop precise abstractions and action operators for use in AI. This enables us to find scalable, algorithm-like plans with branches and loops which can solve problems of unbounded sizes. We describe how this method of abstraction can be effectively used in AI, with compelling results from implementations of two planning algorithms.

Despite a big progress in solving planning problems, more complex problems still remain hard and challenging for existing planners. One of the most promising research directions is exploiting knowledge engineering techniques such as (re)formulating the planning problem to be easier to solve for existing planners. In particular, it is possible to automatically gather knowledge from toy planning problems and exploit this knowledge when solving more complex planning problems. In this paper we propose a method for eliminating some actions from the problem specification that are often useless or may mislead the planners. The method detects if actions are somehow connected with the initial or goal predicates and by using this information we suggest that some actions are not necessary when solving the planning problem. To eliminate these actions we modify the planning domain and hence the method remains independent of used planning system.

Classical planning deals with finding a (shortest) sequence of actions transferring the world from its initial state to a state satisfying the goal condition. Traditional planning systems explore either paths in the state space (state-space planning) or partial plans (plan-space planning). In this paper we show how the ideas from plan-space (partial order) planning can be integrated into state-space (sequential) planning by combining constraint models describing both types of planning. In particular, we extend our existing constraint model for sequential planning by constraints describing satisfaction of open goals. We demonstrate experimentally that this extension pays-off especially when the planning problems become harder.

Conformant planning is the problem of finding a sequence of actions for achieving a goal in the presence of uncertainty in the initial state or action effects. The problem has been approached as a path-finding problem in belief space where good belief representations and heuristics are critical for scaling up. In this work, a different formulation is introduced for conformant problems with deterministic actions where they are automatically converted into classical ones and solved by an off-the-shelf classical planner. The translation maps literals L and sets of assumptions t about the initial situation, into new literals KL/t that represent that L must be true if t is initially true. We lay out a general translation scheme that is sound and establish the conditions under which the translation is also complete. We show that the complexity of the complete translation is exponential in a parameter of the problem called the conformant width, which for most benchmarks is bounded. The planner based on this translation exhibits good performance in comparison with existing planners, and is the basis for T0, the best performing planner in the Conformant Track of the 2006 International Planning Competition.

This paper describes an approach to scheduling under uncertainty that achieves scalability through a coupling of deterministic and probabilistic reasoning. A class of oversubscribed scheduling problems is considered where the goal is to maximize the reward earned by a team of agents in a distributed execution environment. There is uncertainty in both the duration and outcomes of executed activities, and activities are subject to deadlines. To ensure scalability, the approach takes as its starting point an initial deterministic schedule for the agents, computed using expected duration reasoning. This initial agent schedule is probabilistically analyzed to find likely points of failure, and then selectively strengthened based on this analysis. Experimental results obtained in a multi-agent simulation environment demonstrate that coupling probabilistic and deterministic reasoning in this way results in significantly higher rewards than are achieved by relying on deterministic reasoning alone. In the future, the approach will be extended to include probability-driven meta-level management of execution.

We present an implementation of a plan adaptation system, BioPlanner, built for biological pathway prediction across species. BioPlanner formulates a pathway discovery problem as a Hierarchical Task Network (HTN) planning problem and solves it by adapting a plan solution of another well-studied pathway. BioPlanner provides the following functionalities: It automatically builds HTN planning models for a biological pathway domain from the semantic web biological knowledge bases (KBs). It retrieves plan cases from the biological KBs. It generates hypothetical pathways using plan adaptation strategies with the aid of biological domain knowledge. It evaluates the hypothetical plan candidates, ranks them, and recommends the most likely hypotheses to users. It employs an information gathering multi-agent system to capture knowledge from heterogeneous sources to help the hypothetical plan generation process. We utilize BioPlanner to predict Signaling Transduction pathways for Mus musculus, Gallus gallus, and Drosophila melanogaster from Homo sapiens.