Risk-Aware Scheduling throughout Planning and Execution

Scheduling is integral to many real-world logistics problems. It can be as simple as catching the bus in the morning, or as complex as assembling a commercial airliner. While simple applications render scheduling tools trivial, these tools have not been widely adopted for complex scenarios either. The larger the scenario, the greater the temporal uncertainty throughout the system, and many schedulers do not consider the probabilistic uncertainty in actions’ durations. This makes them brittle to temporal disturbances or poor modeling, which incurs high replanning costs and unacceptable risks for the mission-critical nature of large applications. Mitigating such risk is challenging because temporal reasoning is present at every level of planning and execution. Figure 1 diagrams the layers of reasoning for a planning executive to map logistical goals into real-world actions. First, the planner generates an grounded plan. The the scheduler produces a scheduling policy, which the dispatcher then follows to execute the plan’s actions at appropriate times. Although the scheduler is responsible for creating the scheduling policy, the planner and dispatcher still have to be aware of time: The dispatcher is responsible for keeping the plan on schedule. Hence, it needs to monitor actions’ durations to make sure they match what the policy predicts. Likewise, the planner needs to know when the scheduler cannot produce a valid policy for a given plan, and use that information to avoid unschedulable plans. When temporal uncertainty is factored into the scheduling algorithm, the planner and dispatcher must adapt to reason about the consequences of uncertainty as well. Thus, quantifying and managing scheduling risk requires one to apply probabilistic temporal reasoning across the entire planning and execution architecture. If this could be done efficiently, it would remove a significant obstacle to deploying largescale logistics scenarios with confidence.