In many important situations, giving consideration to the time and resource impact of planning decisions made is essential to allow useful plans to be produced. For instance, if the plan is to be executed by an agent with limited battery capacity, the planner must choose actions that respect this limit, including actions to replenish the battery if appropriate. Or, if there is a deadline by which certain goals must be met, then it is important that the actions complete within the desired time-scale.
In this talk, I'll be presenting a flexible planning architecture, POPF, for problems such as these. Its kernel works by constructing a 'Partial Order Plan, Forwards', allowing it to find temporally efficient plans. In its original form, this was combined with a Linear Programming (LP) solver, for solving problems where actions interact with resources continuously over their execution. More recently, as part of SICSA, it has been combined with a Bayesian Network, to estimate the likelihood that a given sequence of plan steps will meet a deadline. To motivate the work, I'll present a case study for each of these.