Robust Execution of Probabilistic STNs

A Probabilistic Simple Temporal Network (PSTN) is a formalism for representing and reasoning about actions subject to temporal constraints, where some action durations may be uncontrollable, modeled using continuous probability density functions. Recent work aims to manage this kind of uncertainty during execution by approximating a PSTN by a Simple Temporal Network with Uncertainty (STNU) (for which well-known execution strategies exist) and using an STNU execution strategy to execute the PSTN, hoping that its probabilistic action durations will not cause any constraint violations. This paper presents significant improvements to the robust execution of PSTNs. Our approach is based on a recent, faster algorithm for finding negative cycles in non-DC STNUs. We also formally prove that many of the constraints included in others' work are unnecessary and that our algorithm can take advantage of a flexible real-time execution algorithm to react to observations of contingent durations that may fall outside the fixed STNU bounds. The paper presents an empirical evaluation of our approach that provides evidence of its effectiveness in robustly executing PSTNs derived from a publicly available benchmark.