The Secret Behind Superior Flight Scheduling Automation

There's an "aha!" moment that happens in nearly every single one of our product demonstrations. Our team renders what a typical flight schedule might look like – a “puckboard”, to some – and with a few clicks, we're able to completely update the board to reflect the most current flying conditions.

Schedulers and operators simply aren't accustomed to this. They're used to having to adjust schedules manually, often spending many hours performing tasks that our technology can accomplish in seconds. Most of our customers have sat through numerous pitches for automation solutions, but the technology almost always falls short of accomplishing the mission. So what is it about the way OpsLab approaches automation that seems to really resonate?

SkySchedule "Weekly Mission Board"

To truly understand this, we first need to understand a week in the life of a military pilot or civilian tasked with maintaining squadron schedules. The following steps are just one representative example, though individual squadrons may vary.

• Before creating or editing schedules, the scheduler has to be aware of the high-level prioritization handed to them by the squadron’s director of operations (DO). Examples: prioritize one student over another, overfly a pilot because of upcoming vacations, schedule missions of a specific type because of seasonal availability of resources.

• Early in the week, the scheduler focuses more on operations of the current day and week. The goal is to check for schedule disruptions, and resolve them while respecting the DO’s priorities and any last-minute changes to the availability of pilots and planes.

• Then, each schedule edit done in the current week can create cascading effects. For example, a syllabus mission removed from today’s schedule may be a prerequisite for a mission that a student has been scheduled to fly later in the week. The scheduler has to watch out for this and ensure that tomorrow’s changes are compatible with today’s updates.

• Finally, mid-week and beyond, the scheduler starts finalizing the schedule for the next week based on what is likely to be executed for the rest of the current week. Here, the scheduler is creating a schedule from scratch by first selecting missions, and then assigning pilots.

All of this is done in a very dynamic environment; DO priorities may change over time, DO short-term priorities can be different from original priorities used to create a schedule, pilots may suddenly unavailable for personal reasons or deployment, instructors may join or leave the squadron, and so on.

Any software solution that truly assists the scheduler must satisfy the following requirements at the minimum:

• Verification: Schedulers must be able to verify necessary conditions required by any schedule, including:avoiding overlapping events for pilots, flying more lines than available aircraft, and students needing adequate training for complex missions.

• Control: When using a software solution for assistance, the scheduler must be able to control which decisions are made, and how they are made. Some schedulers may want the software to do everything (select missions, assign pilots), but others may have more specific needs. When selecting pilots, the schedulers must be able to clearly specify the main drivers for pilot selection (e.g. prioritize combat readiness over upgrades or vice versa).

• Human-in-the-loop: Once the solution has been generated, the scheduler should still have final control for manual overrides. They should also be able to see how their manual changes impact the schedule.

These requirements necessitate a formal framework that can handle at least schedule creation, editing, tracking and reporting. So, how do we create such a framework that can handle the needs of not just 1 scheduler, but multiple wings with frequently changing requirements and resources? Operations Research (OR).

Operations research is an automated decision support system that allows users to input both hard constraints and soft constraints that they fully control. Hard constraints are immutable properties that the system must respect no matter what: every pilot needs 12 hours of overnight rest, maximum 3 flights per week, etc. Soft constraints are more flexible and can be manipulated to fit the needs of a whole squadron, for instance, maintaining consistent schedules for better work-life balance.

Using this framework of hard and soft constraints, our team has been able to develop a system that does not rely on learning or on historical data. We create a system that produces the best possible decision or the best possible schedule or the best possible asset allocation subject to the constraints that you've configured. We're able to respect dozens of scheduling parameters while innovating, creating schedules that are superior to what humans alone can devise.

SkySchedule "Pilot Tracking"

Operations research also offers the key benefit of explainability. It doesn’t suffer from hallucinations or algorithmic biases. Once users set hard constraints, schedules are guaranteed to respect them. Soft constraints are combined into a single mathematical objective function within OR models. This function becomes a consistent scoring system to judge a schedule’s quality. By adjusting this function and trying out schedule changes, users can immediately verify the impact of their changes to schedule quality.

The OR framework we have created is complementary to Machine Learning (ML)-based and Large Language Model (LLM)-based techniques that already aid schedulers. To illustrate with an example, we are prototyping a LLM agent that can process natural language feedback from users, convert it into hard/soft constraints that can change the type of solutions generated by the framework, while ensuring traditional schedule guarantees (e.g. no overlapping events).

Let's bring it back to why this is having an impact with our core customer set: the U.S. military. Warfighters need to be able to constantly innovate their frameworks. They need frameworks that don't mimic and don't rely on the accuracy of the hundreds of disparate data sets. They need operational frameworks that give them fast, accurate, practical outputs. Frameworks that not only show what's happening now, but allow them to predict and forecast.

Consider, too, automation in the context of operational readiness, specifically as it relates to preserving one of our most critical resources: time. The core value of our automation platform is its ability to give back time. In flying squadrons, scheduling time is reduced by more than 500% – from five hours a day to as little as 15 minutes. Many schedulers are also instructors, so this translates to more flight time, more teaching, better mission planning, and yes, more sleep. After all, pilots join the U.S. Air Force to fly, not to wrestle with spreadsheets.

Now extend this basic value proposition out to long-term mission planning. Automation based on operations research not only provides predictive forecasts, but it also demonstrates how those predictions might play out through comprehensive, long-term schedules. In other words, it serves as a powerful "what-if" tool. Commanders can simulate scenarios – like increasing the number of planes or instructors – and instantly see how it affects graduation rates or instructor utilization. This level of scenario planning is unparalleled.

Of course, these principles also extend far beyond aviation. The decision support framework we have built is general and customizable, and it can be applied to any field involving complex scheduling and training – unmanned vehicles, submarines, boats, land vehicles, etc.

In conclusion, any technological success we've achieved so far stems from choosing the right tool: operations research. By allowing for both hard and soft constraints, we’re able to automatically manage deeply complex and dynamic assets. In seconds, we’re able to achieve a level of accuracy that used to take hours. The result is a system that not only plans but also predicts, offering unrivaled support for short-term decisions and long-term scenario planning. If applied correctly, operations research can add value in so many ways – saving time, saving money, reducing wastage, improving efficiency. We’re only just scratching the surface of what automation through operations research can accomplish.