Hector Geffner

Planning as heuristic search

In the AIPS98 Planning Contest, the HSP planner showed that heuristic search planners can be competitive with state-of-the-art Graphplan and SAT planners. Heuristic search planners like HSP transform planning problems into problems of heuristic search by automatically extracting heuristics from Strips encodings. They differ from specialized problem solvers such as those developed for the 24-Puzzle and Rubik’s Cube in that they use a general declarative language for stating problems and a general mechanism for extracting heuristics from these representations. In this paper, we study a family of heuristic search planners that are based on a simple and general heuristic that assumes that action preconditions are independent. The heuristic is then used in the context of best-first and hill-climbing search algorithms, and is tested over a large collection of domains. We then consider variations and extensions such as reversing the direction of the search for speeding node evaluation, and extracting information about propositional invariants for avoiding dead-ends. We analyze the resulting planners, evaluate their performance, and explain when they do best. We also compare the performance of these planners with two state-of-the-art planners, and show that the simplest planner based on a pure best-first search yields the most solid performance over a large set of problems. We also discuss the strengths and limitations of this approach, establish a correspondence between heuristic search planning and Graphplan, and briefly survey recent ideas that can reduce the current gap in performance between general heuristic search planners and specialized solvers.  2001 Elsevier Science B.V. All rights reserved.

Admissible heuristics for optimal planning

The invited speakers at the conference presented some of their latest research and ideas on intelligent planning and execution: Drew McDermott from Yale University gave the first talk, entitled “Bottom-Up Knowledge Representation,” and David Smith from In recent years, AI planning and scheduling has emerged as a technology critical to production management, space systems, the internet, and military applications. The Fifth International Conference on Artificial Intelligence Planning and Scheduling (AIPS2000) was held on 14–17 April 2000 at Breckenridge, Colorado;1 it was colocated with the Seventh International Conference on Principles of Knowledge Representation and Reasoning (KR2000). This conference brought together researchers working in all aspects of problems in planning, scheduling, planning and learning, and plan execution for dealing with complex problems. The format of the conference included paper presentations, invited speakers, panel discussions, workshops, and a planning competition. The conference was cochaired by Steve Chien of the Jet Propulsion Laboratory (JPL) at the California Institute of Technology, Subbarao Kambhampati of Arizona State University, and Craig Knoblock of the University of Southern California Information Sciences Institute, with the proceedings published by AAAI Press (Chien, Kambhampati, and Knoblock 2000). The three workshops were “Analyzing and Exploiting Domain Knowledge for Efficient Planning,” chaired by Maria Fox from University of Durham; “DecisionTheoretic Planning,” chaired by Richard Goodwin from IBM’s T. J. Watson Research Center and Sven Koenig from Georgia Institute of Technology; and “Model-Theoretic Approaches to Planning” by Paolo Traverso from information describing the content of documents or the behavior of programs. Because the described objects need to be processed by a wide variety of programs, designed by many different parties, finding a representation system to describe any content seems to be a daunting challenge. This challenge is similar to the well-known problems in trying to find a “formal theory of everything.” This talk described a more modest bottom-up approach that involves incrementally building small-specialized knowledge representation frameworks for immediate payoffs and facilitates greater payoffs as these small frameworks are linked together. This approach succeeds even if the process never converges to a general-purpose representation. Making it work involves carefully defining notions of when a framework is strictly more expressive than another and what it means to translate expressions within and between frameworks.

Planning as Heuristic Search: New Results

In the recent AIPS98 Planning Competition, the hsp planner, based on a forward state search and a domain-independent heuristic, showed that heuristic search planners can be competitive with state of the art Graphplan and Satisfiability planners. hsp solved more problems than the other planners but it often took more time or produced longer plans. The main bottleneck in hsp is the computation of the heuristic for every new state. This computation may take up to 85% of the processing time. In this paper, we present a solution to this problem that uses a simple change in the direction of the search. The new planner, that we call hspr, is based on the same ideas and heuristic as hsp , but searches backward from the goal rather than forward from the initial state. This allows hspr to compute the heuristic estimates only once. As a result, hspr can produce better plans, often in less time. For example, hspr solves each of the 30 logistics problems from Kautz and Selman in less than 3 seconds. This is two orders of magnitude faster than blackbox. At the same time, in almost all cases, the plans are substantially smaller. hspr is also more robust than hsp as it visits a larger number of states, makes deterministic decisions, and relies on a single adjustable parameter than can be fixed for most domains. hspr, however, is not better than hsp accross all domains and in particular, in the blocks world, hspr fails on some large instances that hsp can solve. We discuss also the relation between hspr and Graphplan, and argue that Graphplan can also be understood as a heuristic search planner with a precise heuristic function and search algorithm.

Branching and pruning: An optimal temporal POCL planner based on constraint programming

A key feature of modern optimal planners such as graphplan and blackbox is their ability to prune large parts of the search space. Previous Partial Order Causal Link (POCL) planners provide an alternative branching scheme but lacking comparable pruning mechanisms do not perform as well. In this paper, a domain-independent formulation of temporal planning based on Constraint Programming is introduced that successfully combines a POCL branching scheme with powerful and sound pruning rules. The key novelty in the formulation is the ability to reason about supports, precedences, and causal links involving actions that are not in the plan. Experiments over a wide range of benchmarks show that the resulting optimal temporal planner is much faster than current ones and is competitive with the best parallel planners in the special case in which actions have all the same duration.

Compiling uncertainty away in conformant planning problems with bounded width

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.

Goal recognition over POMDPs: inferring the intention of a POMDP agent

Plan recognition is the problem of inferring the goals and plans of an agent from partial observations of her behavior. Recently, it has been shown that the problem can be formulated and solved using planners, reducing plan recognition to plan generation. In this work, we extend this model-based approach to plan recognition to the POMDP setting, where actions are stochastic and states are partially observable. The task is to infer a probability distribution over the possible goals of an agent whose behavior results from a POMDP model. The POMDP model is shared between agent and observer except for the true goal of the agent that is hidden to the observer. The observations are action sequences O that may contain gaps as some or even most of the actions done by the agent may not be observed. We show that the posterior goal distribution P(G|O) can be computed from the value function VG(b) over beliefs b generated by the POMDP planner for each possible goal G. Some extensions of the basic framework are discussed, and a number of experiments are reported.

Soft goals can be compiled away

Soft goals extend the classical model of planning with a simple model of preferences. The best plans are then not the ones with least cost but the ones with maximum utility, where the utility of a plan is the sum of the utilities of the soft goals achieved minus the plan cost. Finding plans with high utility appears to involve two linked problems: choosing a subset of soft goals to achieve and finding a low-cost plan to achieve them. New search algorithms and heuristics have been developed for planning with soft goals, and a new track has been introduced in the International Planning Competition (IPC) to test their performance. In this note, we show however that these extensions are not needed: soft goals do not increase the expressive power of the basic model of planning with action costs, as they can easily be compiled away. We apply this compilation to the problems of the net-benefit track of the most recent IPC, and show that optimal and satisficing cost-based planners do better on the compiled problems than optimal and satisficing netbenefit planners on the original problems with explicit soft goals. Furthermore, we show that penalties, or negative preferences expressing conditions to avoid, can also be compiled away using a similar idea.

mGPT: a probabilistic planner based on heuristic search

We describe the version of the GPT planner used in the probabilistic track of the 4th International Planning Competition (IPC-4). This version, called mGPT, solves Markov Decision Processes specified in the PPDDL language by extracting and using different classes of lower bounds along with various heuristic-search algorithms. The lower bounds are extracted from deterministic relaxations where the alternative probabilistic effects of an action are mapped into different, independent, deterministic actions. The heuristic-search algorithms use these lower bounds for focusing the updates and delivering a consistent value function over all states reachable from the initial state and the greedy policy.

A Framework for Reasoning with Defaults

Belief commitment and belief revision are two distinctive characteristics of common sense reasoning which have so far resisted satisfactory formal accounts. Classical logic for instance, cannot accommodate belief revision: new information can only add new theorems. Probability theory, on the other hand, has difficulties in accommodating belief commitment: propositions are believed only to a certain degree which dynamically changes with the acquisition of new information.

The AIPS-98 Planning Competition

In 1998, the international planning community was invited to take part in the first planning competition, hosted by the Artificial Intelligence Planning Systems Conference, to provide a new impetus for empirical evaluation and direct comparison of automatic domain-independent planning systems. This article describes the systems that competed in the event, examines the results, and considers some of the implications for the future of the field.