Planning under partial observability is one of the most significant and challenging planning problems. It has been

The Model Based Planner (MBP) is a system for planning in non-deterministic domains. It can generate plans automatically to solve various planning problems, like conformant planning, planning under partial observability, and planning for temporally extended goals. Moreover, MBP can validate plans, and offers a variety of simulation functionalities for plans and domains. MBP is based on Symbolic Model Checking techniques, and Binary Decision Diagrams (BDDs), that provide a practical solution to the problem of dealing with the large size of realistic planning problems. Experimental analysis in the course of the last years has shown MBP to be state-of-the-art in planning for nondeterministic domains. The demo aims at showing MBP’s array of functionalities for plan generation, validation and simulation over an increasingly complex navigation problem.

Real-time heuristic search methods interleave planning and plan executions and plan only in the part of the domain around the current state of the agents. This is the part of the domain that is immediately relevant for the agents in their current situation. So far, real-time heuristic search methods have mostly been applied to deterministic planning tasks. In this article, we argue that real-time heuristic search methods can efficiently solve nondeterministic planning tasks. Planning in nondeterministic domains can be time-consuming due to the many contingencies. However, real-time heuristic search methods allow agents to gather information early. This information can be used to resolve some of the uncertainty caused by nondeterminism and thus reduce the amount of planning done for unencountered situations. To this end, we introduce Min-Max Learning Real-Time A * (Min-Max LRTA*), a real-time heuristic search method that generalizes Korf’s LRTA * to nondeterministic domains. Min-Max LRTA * has the following advantages: First, different from the many existing ad-hoc planning methods that interleave planning and plan executions, it has a solid theoretical foundation and is domain independent. Second, it allows for fine-grained control over how much planning to do between plan executions. Third, it can use heuristic knowledge to guide planning which can reduce planning time without increasing the plan-execution time. Fourth, it can be interrupted at any state and resume execution at a different

Planning under partial observability is one of the most significant and challenging planning problems. It combines

The last decade has witnessed a dramatic progress in the variety and performance of techniques and tools for classical planning. The existence of a de-facto standard modeling language for classical planning, PDDL, has played a relevant role in this process. PDDL has fostered information sharing and data exchange in the planning community, and has made international classical planning competitions possible. At the same time,

Planning in non-deterministic domains yields both conceptual and practical diculties. From the conceptual point of view, dierent notions of planning problems can be devised: for instance, a plan might either guarantee goal achievement, or just have some chances of success. From the practical point of view, the problem is to devise for algorithms that can deal eectively with large state spaces. In this paper, we tackle planning in non-deterministic domains by addressing conceptual and practical problems. We formally characterize dierent planning problems, where solutions have a chance of success (\weak planning"), are guaranteed to achieve the goal (\strong planning"), or achieve the goal with iterative trial-and-error strategies (\strong cyclic planning"). In strong cyclic planning, all the executions associated with the solution plan always have a possibility of terminating and, when they do, they are guaranteed to achieve the goal.

A learning to coordinate paradigm was rst introduced in Formal Learning Theory by [MO99] using the tools of recursion theory. In this paper, we advance and discuss a rst-order paradigm of coordination|we call this paradigm of model-coordination. The paradigm is shown to extend Montagna and Osherson's paradigm of learning to coordinate, in the sense that Montagna and Osherson's binary players coordinate if and only if their rstorder equivalent agents model-coordinate. An important dierence between our paradigm and that proposed by [MO99] is that in our paradigm agents' preferences and beliefs can be modelled.

In this paper, we explore an architecture, called K-Trek, that enables mobile users to travel across knowledge distributed over a large geographical area (ranging from large public buildings to a national park). Our aim is providing, distributing, and enriching the environment with location-sensitive information for use by agents on board of mobile and static devices. Local interactions among KTrek devices and the distribution of information in the larger environment adopt some typical peer-to-peer patterns and techniques. We introduce the architecture, discuss some of its potential knowledge management applications, and present a few experimental results obtained with simulation.

Interleaving planning and execution is the practical alternative to the problem of planning off-line with large state spaces. While planning via symbolic model checking has been extensively studied for off-line planning, no framework for interleaving it with execution has been ever devised. In this paper, we extend planning via symbolic model checking with the ability of interleaving planning and execution in the case of nondeterministic domains and partial observability, one of the most challenging and complex planning problems. We build

Robots often have to navigate robustly despite incomplete information about the terrain or their location in the terrain. In this case, they often use greedy methods to make planning tractable. In this paper, we analyze two such robot-navigation methods. The first