Diagnosing Dependent Action Delays in Temporal Multiagent Plans

Diagnosis of Temporal Multiagent Plans (TMAPs) aims at identifying the causes of delays in achieving the plan goals. So far, approaches to TMAP diagno- sis have relied on an assumption that might not hold in many practical domains: action delays are independent of one another. In this paper we relax this assump- tion by allowing (indirect) dependencies among action delays. The diagnosis of a given TMAP is inferred by exploiting a qualitative Bayesian Network (BN), through which dependencies among actions delays, even performed by different agents, are captured. The BN, used to compute the heuristic function, drives a standard A* search, which finds all the most plausible explanations. Results of a preliminary experimental analysis show that the proposed Bayesian-based heuristic function is feasible.