In this article, we describe a possibilistic/probabilistic conditional planner called PTLplan. Being inspired by Bacchus and Kabanza's TLplan, PTLplan is a progressive planner that uses strategic knowledge encoded in a temporal logic to reduce its search space. Actions effects and sensing can be context dependent and uncertain, and the information the planning agent has at each point in time is represented as a set of situations with associated possibilities or probabilities. Besides presenting the planner itself -- its representation of actions and plans, and its algorithm -- we also provide some promising data from performance tests.

This work is concerned with the automatic inference of generic types from STRIPS planning domain descriptions. Generic types are higher order types allowing the partition of domains (and components of domains) into different domain classes, including the commonly occurring transportation domain class. We show how the generic type structure of domains can be exploited to increase planner efficiency. We have focussed so far on the generic types typical of transportation domains, but intend to go on to characterise, and identify examples of, other domain classes such as construction domains. One of the most interesting properties of the work described here is that domains which would not be recognised, by the human, as transportation domains can turn out to have an underlying transportation character which can be exploited by the application of heuristics suited to standard transportation domains. We illustrate this by considering both standard transportation domains (such as Logistics) and non-standard ones (the PaintWall domain presented in this paper). The analyses described here are completely planner-independent and contribute to an increasing collection of pre-plannin 9 analysis tools which help to increase performance of planners by decomposing and understanding the structures of planning problems before planners are applied.

The Model Checking Integrated Planning System (MIPS) has shown distinguished performance in the second and third international planning competitions. With its object-oriented framework architecture MIPS clearly separates the portfolio of explicit and symbolic heuristic search exploration algorithms from different on-line and off-line computed estimates and from the grounded planning problem representation.

This paper proposes the problem of supply restoration in faulty power distribution systems as a benchmark for planning under uncertainty. This benchmark, which is derived from a significant realworld case, is both simple to understand and easily scalable. The goal is to reconfigure the distribution network to resupply a maximum of consumers a#ected by the faults. Due to sensor and actuator uncertainty, the location of the faulty areas and the current network configuration are only partially observable. This makes the problem very challenging. 1 Motivation The use of poor benchmarks for planning under uncertainty has often been pointed out as detrimental to the impact of the field on the wider community. Except for a few testbeds in robot navigation, see e.g. [6], we are still confined to purely artificial problems ranging from escaping the tiger behind the door to making an omelette. While well-understood toy problems are definitely useful in explaining performance di#erence...

Despite the long history of research in using machine learning to speed-up state-space planning, the techniques that have been developed are not yet in widespread use in practical planning systems. One limiting factor is that traditional domain-independent planning systems scale so poorly that extensive learned control knowledge is required to raise their performance to an acceptable level. Therefore, work in this area has focused on learning large numbers of control rules that are specific to the details of the underlying planning algorithms, which can be extremely costly. In recent years, a new generation of planning systems with much improved speed and scalability has become available. These systems formulate planning as solving a large constraint satisfaction problem. This formulation opens up the possibility that domain-specific control knowledge can be added to the planner in a purely declarative manner via a set of additional constraints. In this paper we present the first positive results on automatically acquiring such high-level, declarative constraints using machine learning techniques. In particular, we will show that a new heuristic method for generating training examples together with a rule induction algorithm can learn useful control rules in a variety of domains. Only a small number of rules are needed to reduce solution times by two orders of magnitude or more on larger problems, training times are short, and the learned rules can be exported to other planning systems.

We present a new forward chaining planner, TALplanner, based on ideas developed by Bacchus [5] and Kabanza [11], where domain-dependent search control knowledge represented as temporal formulas is used to effectively control forward chaining. Instead of using a linear modal tense logic as with Bacchus and Kabanza, we use TAL, a narrative-based linear temporal logic used for reasoning about action and change in incompletely specified dynamic environments. Two versions of TALplanner are considered, TALplan/modal which is based on the use of emulated modal formulas and a progression algorithm, and TALplan/non-modal which uses neither modal formulas nor a progression algorithm. For both versions of TALplanner and for all tested domains, TALplanner is shown to be considerably faster and requires less memory. The TAL versions also permit the representation of durative actions with internal state. In proceedings: 6th Int'l Workshop on Temporal Representation and Reasoning (TIME-99), IEEE, 19...

We describe and evaluate a system for learning domainspecific control knowledge. In particular, given a planning domain, the goal is to output a control policy that performs well on “long random walk ” problem distributions. The system is based on viewing planning domains as very large Markov decision processes and then applying a recent variant of approximate policy iteration that is bootstrapped with a new technique based on random walks. We evaluate the system on the AIPS-2000 planning domains (among others) and show that often the learned policies perform well on problems drawn from the long–random-walk distribution. In addition, we show that these policies often perform well on the original problem distributions from the domains involved. Our evaluation also uncovers limitations of our current system that point to future challenges.

Domain-independent planning is a hard combinatorial problem. Taking into account plan quality makes the task even more difficult. This article introduces Planning by Rewriting (PbR), a new paradigm for efficient high-quality domain-independent planning. PbR exploits declarative plan-rewriting rules and efficient local search techniques to transform an easy-to-generate, but possibly suboptimal, initial plan into a high-quality plan. In addition to addressing the issues of planning efficiency and plan quality, this framework offers a new anytime planning algorithm. We have implemented this planner and applied it to several existing domains. The experimental results show that the PbR approach provides significant savings in planning effort while generating high-quality plans.

We describe a forward reasoning planner for open worlds that uses domain specific information for pruning its search space, as suggested by (Bacchus & Kabanza 1996; 2000). The planner is written in the situation calculus-based programming language GOLOG, and it uses a situation calculus axiomatization of the application domain. Given a sentence oe to prove, the planner regresses it to an equivalent sentence oe 0 about the initial situation, then invokes a theorem prover to determine whether the initial database entails oe 0 and hence oe. We describe two approaches to this theorem proving task, one based on compiling the initial database to prime implicate form, the other based on Relsat, a Davis/Putnam-based procedure. Finally, we report on our experiments with open world planning based on both these approaches to the theorem proving task.

In this paper, we present a general technique for taking forward-chaining planners for deterministic domains (e.g., HSP, TLPlan, TALplanner, and SHOP2) and adapting them to work in nondeterministic domains. Our results