We present a new algorithm for probabilistic planning with no observability. Our algorithm, called Probabilistic-FF, extends the heuristic forward-search machinery of Conformant-FF to problems with probabilistic uncertainty about both the initial state and action effects. Specifically, Probabilistic-FF combines Conformant-FF’s techniques with a powerful machinery for weighted model counting in (weighted) CNFs, serving to elegantly define both the search space and the heuristic function. Our evaluation of Probabilistic-FF on several probabilistic domains shows an unprecedented, several orders of magnitude improvement over previous results in this area.

We present a case study in confronting the GPT generalpurpose planner with the challenging power supply restoration (PSR) benchmark for contingent planning. PSR is derived from a real-world problem, and the difculty of modeling and solving it contrasts with that of the purely articial benchmarks commonly used in the literature. This confrontation leads us to improve general techniques for contingent planning, to provide a PDDL-syle encoding of PSR which we hope to see used in planning competitions, and to report the rst results on generating optimal policies for PSR.

. In this paper, we propose new agent architecture which adapts its own behavior by avoiding actions which are predicted to be failure. We name this agent inductive learning agent (ILA). This agent consists of ve parts: Observer, Planner, Checker, Actor, and Learner. These parts use rst-order formalism and inductive logic programming (ILP) to acquire rules to predict. Adapting behavior of ILA works as follows: (1) Collect examples of actions. (2) Classify the examples. (3) Acquire prediction rules using ILP. (4) Behave under the prediction rules. We have implemented it in soccer using parts of the latest RoboCup competition champion CMUnited-99 and an ILP system Progol. We have conrmed that agents could acquire prediction rules and could adapt their behavior using the rules. 1 Introduction Researchers have studied on machine learning in the toy world and tried to bring them to the real world. However, it gradually became clear that some of these technologies developed i...

Abstract not found

In this article, we describe a possibilistic/probabilistic conditional planner called PTLplan, and how this planner can be integrated with a behavior-based fuzzy control system called the Thinking Cap in order to execute the generated plans. 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 ' eects and sensing can be context dependent and uncertain, and the resulting plans may contain conditional branches. When these plans are executed by the control system, they are transformed into B-plans which essentially are combinations of fuzzy behaviors to be executed in dierent contexts.

We present a new algorithm for probabilistic planning with no observability. Our algorithm, called Probabilistic-FF, extends the heuristic forward-search machinery of Conformant-FF to problems with probabilistic uncertainty about both the initial state and action effects. Specifically, Probabilistic-FF combines Conformant-FF’s techniques with a powerful machinery for weighted model counting in (weighted) CNFs, serving to elegantly define both the search space and the heuristic function. Our evaluation of Probabilistic-FF shows its fine scalability in a range of probabilistic domains, constituting a several orders of magnitude improvement over previous results in this area. We use a problematic case to point out the main open issue to be addressed by further research. 1.

We describe Probapop, a partial-order probabilistic planning system. Probapop is a blind (conformant) planner that finds plans for domains involving probabilistic actions but no observability. The Probapop implementation is based on Vhpop, a partial-order deterministic planner written in C++. The Probapop algorithm uses plan graph based heuristics for selecting a plan from the search queue, and probabilistic assessment heuristics for selecting a condition whose probability can be increased.

Abstract not found

In this paper we report on the extension of the classical HTN planner SHOP to plan in partially observable domains with uncertainty. Our algorithm PC-SHOP uses belief states to handle situations involving incomplete and uncertain information about the state of the world. Sensing and acting are integrated in the primitive actions through the use of a stochastic model. PC-SHOP is showed to scale up well compared to some of the state-of-the-art planners. We outline the main characteristics of the algorithm, and present performance results on some problems found in the literature.

Probabilistic planning is an inherently multi-objective problem where plans must trade-off probability of goal satisfaction with expected plan cost. To date, probabilistic plan synthesis algorithms have focussed on single objective formulations that bound one of the objectives by making some unnatural assumptions. We show that a multi-objective formulation is not only needed, but also enables us to (i) generate Pareto sets of plans, (ii) use recent advances in probabilistic planning reachability heuristics, and (iii) elegantly solve limited contingency planning problems. We extend LAO ∗ to its multi-objective counterpart MOLAO ∗ , and discuss a number of speed-up techniques that form the basis for a state of the art conditional probabilistic planner. 1