Uncertainty-Aware Drone Dispatch: Recording-Energy Prediction and Charging-Constrained Simulation

Imagine a fleet of drones waiting at a depot, ready to deliver packages. Each drone carries a battery, and the person in charge must decide which drone to send for which job. The decision is a tightrope walk. If the dispatcher sends a drone that does not have quite enough power, the drone might run out of energy mid-flight, leaving the package stranded and the battery dead. But if the dispatcher is too cautious and refuses to send drones that actually have enough power, the depot sits idle, and customers wait longer than necessary. The core challenge is knowing exactly how much energy a specific flight will consume before the drone ever leaves the ground. This is not just about guessing; it is about measuring the invisible cost of uncertainty. When a flight plan is created, it includes a speed, a weight, and a height, but it does not include the exact amount of electricity the motors will burn. That number only becomes clear after the flight is finished and the data is downloaded.

A researcher named Samer Abaddi set out to solve this puzzle by looking at real flight data from a specific drone, the DJI Matrice 100, which was flown on a common route in Pennsylvania. He gathered records from 209 separate flights, but after checking the data for errors and consistency, he focused on 181 recordings that followed a strict set of rules. These flights carried different weights, flew at different speeds, and hovered at different altitudes. The goal was to build a system that could look at a planned flight and predict the energy it would need, while also creating a safety margin to account for the fact that no prediction is perfect. The researcher used a statistical method called ridge regression, which is a way of finding the best mathematical relationship between the flight settings and the energy used, without letting the model get confused by small, random fluctuations in the data. He then tested this model on new flights it had never seen before to see how close the predictions were to the reality.

The results showed that the model was remarkably accurate. When predicting the total energy for a flight, the average difference between the prediction and the actual energy used was less than one watt-hour, a very small amount for a drone battery. However, accuracy alone is not enough for a dispatcher who cannot afford to lose a drone. The researcher needed to know how much extra power to add to the prediction to be sure the drone would make it. He developed a rule that looked at the worst-case errors from a set of past flights and added a buffer to cover them. This created an "upper estimate," a number that was slightly higher than the most likely energy use, designed to ensure the drone would not run out of power. When this rule was applied to the reserved test flights, every single one of them used less energy than the upper estimate predicted. The safety margin worked perfectly for the data at hand.

To see how this safety margin affected real-world operations, the researcher built a computer simulation of a drone depot. He created a virtual world where requests for deliveries arrived randomly, and two different rules competed to see which one could handle the most work. One rule used the simple, most-likely energy prediction, while the other used the cautious upper estimate with the safety buffer. The simulation ran through thousands of scenarios, mimicking a busy day at a depot with two drones and one charger. The findings revealed a clear trade-off. The cautious rule, which used the upper estimate, successfully prevented the drones from ever dipping below their required reserve power. In the main test scenario, the cautious rule had zero instances where a drone ran out of reserve energy, whereas the simpler rule had five such incidents. However, this safety came at a cost. Because the cautious rule was stricter, it turned down more requests. In the main scenario, it completed 22 fewer deliveries than the simpler rule. The drones were safer, but the depot served fewer customers.

The study also explored what happened when conditions changed, such as when the drones had to carry heavier loads or when the charging station was slower. In every single one of the eighteen different scenarios tested, the cautious rule completed fewer requests than the simple rule. It also reduced the number of times the drones broke their reserve limits in sixteen of those scenarios. In two scenarios, the number of reserve breaches was the same for both rules, but the cautious rule still delivered fewer packages. This suggests that while being overly cautious protects the battery, it consistently reduces the overall efficiency of the depot. The researcher noted that the simulation assumed a specific way of charging the batteries and a specific type of route, so the exact numbers might change in a different environment, but the pattern of trade-off remained steady.

To help others understand these choices, the researcher created an interactive website where anyone can see the data for themselves. A user can select a specific flight plan, change the amount of battery charge available, and see whether a drone would be sent out under the simple rule or the cautious rule. The tool shows a narrow range of battery levels where the two rules disagree: in this zone, the simple rule says "go," while the cautious rule says "wait." This visualizes the exact point where the desire for safety clashes with the desire for speed. The research concludes that there is no perfect solution that maximizes both safety and service. A depot can choose to have fewer reserve breaches by completing fewer deliveries, or it can choose to deliver more packages while accepting a higher risk of running out of power. The value of this work is that it makes that choice visible and measurable, turning a vague feeling of risk into a concrete number that operators can use to decide how much safety they are willing to pay for in lost deliveries.