Cees Witteveen

Plan coordination by revision in collective agent based systems

Abstract In order to model plan coordination behavior of agents we develop a simple framework for representing plans, resources and goals of agents. Plans are represented as directed acyclic graphs of skills and resources that, given adequate initial resources, can realize special resources, called goals. Given the storage costs of resources, application costs of skills, and values of goals, it is possible to reason about the profits of a plan for an agent. We then model two forms of plan coordination behavior between two agents, viz. fusion , aiming at the maximization of the total yield of the agents involved, and collaboration , which aims at the maximization of the individual yield of each agent. We argue how both forms of cooperation can be seen as iterative plan revision processes. We also present efficient polynomial algorithms for agent plan fusion and collaboration that are based on this idea of iterative plan revision. Both the framework and the fusion algorithm will be illustrated by an example from the field of transportation, where agents are transportation companies.

A Resource Logic for Multi-Agent Plan Merging

In a multi-agent system, agents are carrying out certain tasks by executing plans. Consequently, the problem of finding a plan, given a certain goal, has been given a lot of attention in the literature. Instead of concentrating on this problem, the focus of this paper is on cooperation between agents which already have constructed plans for their goals. By cooperating, agents might reduce the number of actions they have to perform in order to fulfill their goals. The key idea is that in carrying out a plan an agent possibly produces side products that can be used as resources by other agents. As a result, an other agent can discard some of its planned actions. This process of exchanging products, called plan merging, results in distributed plans in which agents become dependent on each other, but are able to attain their goals more efficiently. In order to model this kind of cooperation, a new formalism is developed in which side products are modeled explicitly. The formalism is a resource logic based on the notions of resource, skill, goal, and service. Starting with some resources, an agent can perform a number of skills in order to produce other resources which suffice to achieve some given goals. Here, a skill is an elementary production process taking as inputs resources satisfying certain constraints. A service is a serial or parallel composition of skills acting as a program. An operational semantics is developed for these services as programs. Using this formalism, an algorithm for plan merging is developed, which is anytime and runs in polynomial time. Furthermore, a variant of this algorithm is proposed that handles the exchange of resources in a more flexible way. The ideas in the paper will be illustrated by an example from public transportation.

Models and methods for plan diagnosis

We consider a model-based diagnosis approach to the diagnosis of plans. Here, a plan performed by some agent(s) is considered as a system to be diagnosed. We introduce a simple formal model of plans and plan execution where it is assumed that the execution of a plan can be monitored by making partial observations of plan states. These observed states are used to compare them with states predicted based on (normal) plan execution. Deviations between observed and predicted states can be explained by qualifying some plan steps in the plan as behaving abnormally. A diagnosis is a subset of plan steps qualified as abnormal that can be used to restore the compatibility between the predicted and the observed partial state. Besides minimum and subset minimal diagnoses, we argue that in plan-based diagnosis maximum informative diagnoses should be considered as preferred diagnoses, too. The latter ones are diagnoses that make the strongest predictions with respect to partial states to be observed in the future. We show that in contrast to minimum diagnoses, finding a (subset minimal) maximum informative diagnosis can be achieved in polynomial time. Finally, we show how these diagnoses can be found efficiently if the plan is distributed over a number of agents.

A likelihood-ratio test for identifying probabilistic deterministic real-time automata from positive data

We adapt an algorithm (RTI) for identifying (learning) a deterministic real-time automaton (DRTA) to the setting of positive timed strings (or time-stamped event sequences). An DRTA can be seen as a deterministic finite state automaton (DFA) with time constraints. Because DRTAs model time using numbers, they can be exponentially more compact than equivalent DFA models that model time using states. We use a new likelihood-ratio statistical test for checking consistency in the RTI algorithm. The result is the RTI+ algorithm, which stands for real-time identification from positive data. RTI+ is an efficient algorithm for identifying DRTAs from positive data. We show using artificial data that RTI+ is capable of identifying sufficiently large DRTAs in order to identify real-world real-time systems.

An analysis of multi-agent diagnosis

This paper analyzes the use of a Multi-Agent System for Model-Based Diagnosis. In a large dynamical system, it is often infeasible or even impossible to maintain a model of the whole system. Instead, several incomplete models of the system have to be used to establish a diagnosis and to detect possible faults. These models may also be physically distributed.A Multi-Agent System of diagnostic agents may offer solutions for establishing a global diagnosis. If we use a separate agent for each incomplete model of the system, establishing a global diagnosis becomes a problem of cooperation and negotiation between the diagnostic agents. This raises the question whether `a set of diagnostic agents, each having an incomplete model of the system, can (efficiently) determine the same global diagnosis as an ideal single diagnostic agent having the combined knowledge of these agents?

Coordinating Self-interested Planning Agents

We consider planning problems where a number of non-cooperative agents have to work on a joint problem. Such problems consist in completing a set of interdependent, hierarchically ordered tasks. Each agent is assigned a subset of tasks to perform for which it has to construct a plan. Since the agents are non-cooperative, they insist on planning independently and do not want to revise their individual plans when the joint plan has to be assembled from the individual plans. We present a general formal framework to study some computational aspects of this non-cooperative coordination problem and we establish some complexity results to identify some of the factors that contribute to the complexity of this problem. Finally, we illustrate our approach with an application to coordination in multi-modal logistic planning.

Skeptical reason maintenance and belief revision

Abstract The skeptical semantics is a three-valued semantics for reason maintenance based on an extension of the well-known two-valued grounded or stable model semantics. Unlike the latter, however, the skeptical semantics has a computationally attractive feature: the skeptical model can be computed in O(n2) time. The skeptical semantics can also be used to give a better account of the belief revision problem in reason maintenance. A recent logical reconstruction of dependency-directed backtracking (DDB) offers the possibility to represent different DDB strategies by different extensions of a reason maintenance system. Given a reason maintenance system D, we can distinguish a class of extensions, representing all possible DDB strategies for D. We will prove that within this class there exists a unique extension whose skeptical model can be used as a canonical, information-minimal belief revision model. This skeptical belief revision model has some important advantages: 1. (1) The arbitrariness of solutions found by classical dependency-directed backtracking methods can be avoided. 2. (2) The semantics guarantees a (tractable) incremental updating method, and this method satisfies—contrary to standard belief revision techniques—a weak rationality postulate ensuring the minimality of performed changes. 3. (3) Skeptical belief revision is a complete belief revision strategy and is easy to compute, having a worst-case complexity O(n3).

Context-aware route planning

In context-aware route planning, there is a set of transportation agents each with a start and destination location on a shared infrastructure. Each agent wants to find a shortest-time route plan without colliding with any of the other agents, or ending up in a deadlock situation. We present a single-agent route planning algorithm that is both optimal and conflict-free. We also present a set of experiments that compare our algorithm to finding a conflict-free schedule along a fixed path. In particular, we will compare our algorithm to the approach where the shortest conflict-free schedule is chosen along one of k shortest paths. Although neither approach can guarantee optimality with regard to the total set of agent route plans -- and indeed examples can be constructed to show that either approach can outperform the other -- our experiments show that our approach consistently outperforms fixed-path scheduling.

Primary and secondary diagnosis of multi-agent plan execution

Diagnosis of plan failures is an important subject in both single- and multi-agent planning. Plan diagnosis can be used to deal with plan failures in three ways: (i) to provide information necessary for the adjustment of the current plan or for the development of a new plan, (ii) to point out which equipment and/or agents should be repaired or adjusted to avoid further violation of the plan execution, and (iii) to identify the agents responsible for plan-execution failures. We introduce two general types of plan diagnosis: primary plan diagnosis identifying the incorrect or failed execution of actions, and secondary plan diagnosis that identifies the underlying causes of the faulty actions. Furthermore, three special cases of secondary plan diagnosis are distinguished, namely agent diagnosis, equipment diagnosis and environment diagnosis.

A resource based framework for planning and replanning

We discuss a rigorous unifying framework for both planning and replanning, extending an existing logic-based approach to resource-based planning. The primitive concepts in this action resource framework (ARF) are actions and resources. Actions consume and produce resources. Plans are structures composed of actions, resource facts and an explicit dependency function specifying their interrelationships. In this framework, both planning and replanning are conceived as plan transformation processes accomplished by applying sequences of operations on plans. For this, we introduce operators for plan transformation and define the concept of a plan library. Using a refinement planning template, we show how some existing (re)planning methods and heuristics can be described as special cases of this framework. The advantage of the framework is that it offers a unifying view on planning and replanning.