Explaining interdependent action delays in multiagent plans execution

Abstract

In this paper we address the problem of diagnosing the execution of Multiagent Plans with interdependent action delays. To this end, we map our problem to the Model-Based Diagnosis setting, and solve it by devising a novel modeling and reasoning method to infer preferred diagnoses based on partial observation of the start and end times of plan actions. Interestingly, we show that the kind of problem we address can be seen as an extension to the well known disjunctive temporal problem with preferences, augmented with a (qualitative) Bayesian network that models dependencies among action delays. An extensive set of tests performed with a prototype implementation on two different problem domains proves the feasibility of the proposed methodology.

Scheduling with Structured Preferences

“…A CDI-VDTP is an extension to VDTPs in which the evaluation structure S and mapping ϕ are substituted by a Utility Difference Network G, and a utility function u over G. DRIVE slow [15,25] fast [8,13] DRILL deep [10,13] shallow [5,7] COMM chn-1 [10,15] chn-2 [7,10] ANALYZE test-1 [7,9] test-2 [4,5] test-3 [3,4] More formally, a CDI-VDTP is a tuple X, C, G, u , where X and C are as in a standard VDTP; whereas, G = A, E is a directed acyclic graph representing a Utility Difference Network such that:…”

“…Due to space reasons, we just give some insights about the complexity of the proposed solve-CDI-VDTP algorithm; a more detailed analysis of a similar algorithmic approach applied to multi-agent diagnosis can be found in [8]. First of all, we note that the algorithm adopts a recursive strategy for exploring the search space, whose size is bounded by the size of the largest attribute domain, let say D max , and by the number of attributes |A|, namely by the upper bound D max |A| (note, however, that the exploration of the whole search space is very unlikely to occur, since this would require that function consistent never prunes the domains of the attributes).…”

“…The two main sub-functions of solve-CDI-VDTP, namely consistent and MPC, can in principle hide further significant computational cost. It is possible to show that the former is polynomial in |A| as the consistency check can be reduced to a number of invocations of checks over a Simple Temporal Network (STN) proportional to |A| (see [8]). On the other hand, MPC is more complex, as we have seen, since its computation mirrors the computation of the MPE in Bayesian networks, which, as pointed out by Park and Darwiche [13], can be computed in space and time complexity exponential in the width of a given order of the BN nodes, and such a width is itself O(|A|).…”

“…{roberto.micalizio,gianluca.torta}@unito.it sive enough to model scheduling problems [12], as well as other problems of interest in AI (e.g., diagnosis [14,7]). More recently, the research has been focused on how to address temporal preferences (i.e., soft constraints).…”

A Scheduling Tool for Conditionally Independent Temporal Preferences

“…In particular, the notion of Disjunctive Temporal Problems (DTPs) [2] has been introduced to overcome the limits of Simple Temporal Problems (STPs) [1] by enabling the specification of temporal constraints consisting of disjuncts, each of which represents a temporal interval within which legal solutions can be found. This class of problems is expressive enough to model scheduling problems [3], as well as other problems of interest in AI (e.g., diagnosis [4], [5]). In [6], [7] the notion of flexible schedule is introduced.…”