AI in Aviation Case Study: Flight Scheduling Using Digital Twins and Reinforcement Learning

Our client, one of the world’s foremost aerospace companies, continuously seeks to innovate in flight planning. This complex domain involves real-time conditions, regulatory constraints, and evolving airline demands. Our client provides solutions that simplify and optimize flight planning and dispatch operations worldwide. However, traditional flight planning often relies on time-consuming manual adjustments and static optimizers that cannot handle the dynamic nature of modern aviation.

For example, in late December 2022 Southwest Airline cancelled 50% of their flights, ostensibly due to bad weather. It took days to recover, with pilots and flight attendants reporting that they were still on hold when they woke the next morning. After operations had been restored, a post-mortem established that the bottleneck was an inability to reschedule flights, crews and pilots.

To address these challenges, our client collaborated with Winder.AI, an AI agency known for its expertise in reinforcement learning (RL). The goal was to build a digital twin of airline traffic, develop a robust simulator, and train RL agents to determine how best to reschedule flights. This approach aims to improve the speed and efficiency of airline operations and further integrate with the capabilities of our client’s software.

The Research Project

Objectives

The project involved two primary phases. First, Winder.AI built a simulation environment that leveraged a digital twin of real airline traffic. This simulator allowed the team to run scenario-based analyses, which made it easier to explore different flight scheduling disruptions. Second, reinforcement learning agents were introduced to learn how to reschedule flights in the simulated environment. These agents worked to reduce the reliance on manual oversight and slow legacy optimizers.

Both phases aimed to lay the groundwork for a next generation flight planning solution. This would potentially revolutionize how airlines approach scheduling by automating complex decisions and incorporating data-driven insights that adapt to constantly changing conditions.

Why Winder.AI

Our client selected Winder.AI because of its deep experience in reinforcement learning. The consultancy’s team has executed successful commercial RL projects and are authors of O’Reilly’s Reinforcement Learning. Multiple client teams contributed aviation domain expertise that complemented Winder.AI’s data science skills. This collaboration allowed for new research directions and practical proofs of concept in flight scheduling.

Approach and Methodology

Digital Twin and Simulator

Winder.AI collaborated with our client to identify crucial parameters that define airline traffic flows, including turnaround times and aircraft characteristics. Although real operational data was limited, the team leveraged digital twins and open-source frameworks to develop a realistic simulation environment. They refined this simulator in iterative steps, incorporating feedback from subject matter experts.

Reinforcement Learning Agents

Once the simulator was ready, Winder.AI configured RL agents to interact with the environment. The agents were trained to adjust flight schedules, guided by reward structures that reflected real-world goals such as minimizing delays and disruptions. As the agents explored different strategies, they learned to balance multiple priorities, including on-time performance, operational constraints, and overall resource utilization.

Key Challenges

One of the major challenges was gaining access to comprehensive real-world data. Because flight scheduling can involve sensitive commercial information, the project had to rely heavily on digital twins and simulations of the environment. Building a robust simulator required balancing realism with computational feasibility. It was also critical to design RL training procedures that could handle varied scenarios and still produce meaningful scheduling strategies.

Research Findings and Potential Impact

Although a high-fidelity simulation is difficult to create, early results confirmed that it is worthwhile to attempt. Being able to run realistic “what-if” scenarios opens up possibilities for training advanced agents and uncovering insights about operational bottlenecks. Reinforcement learning showed promise for outperforming slow, rule-based optimizers when adjusting flight schedules. This means airlines can potentially shift away from manual scheduling and toward automated, adaptive decision-making.

Looking forward, these findings could significantly impact an airline’s flight planning strategy. Integrating simulation-driven RL methods may lead to faster, more robust software, offering a competitive edge to both our client and the airlines it serves.

Strategic and Business Implications

This research project illustrates a major leap toward modernizing how the aviation industry manages flight scheduling. Current workflows often require intensive manual effort and they rely on static tools that struggle with rapidly changing operational conditions. By embedding RL-driven optimization and simulation insights into our client’s software, they can differentiate in the market and reshape best practices across the industry.

The investment in simulation is also likely to improve the company’s broader flight planning ecosystem. This includes the ability to test improvements or new operational concepts without disrupting live operations.

Conclusion and Next Steps

Through its partnership with Winder.AI, our client is moving toward a more intelligent and adaptive flight planning system, which leverages a digital twin for simulations and employs reinforcement learning agents for optimization. Although further refinements and expanded data access are essential, the early success of this approach suggests a promising roadmap for our client’s flight planning portfolio.

Future stages of this initiative may involve running more extensive simulation scenarios, refining agent algorithms, and integrating proven features into commercial flight planning products. By continuing to invest in simulation and RL, businesses can maintain a leadership role in shaping the future of airline operations.

Learn More or Get in Touch

Explore the Winder.AI website to see how their expertise in reinforcement learning, simulation, and data science can transform your organization’s scheduling and optimization challenges.