While drivers can count on traffic lights to keep vehicles flowing, pilots still rely heavily on human judgement to do likewise. For air traffic controllers (ATCOs), making such calls is no easy feat, especially in an airport’s immediate surroundings. At this busy aerial ‘intersection’—known as the Terminal Manoeuvring Area (TMA)—ATCOs must ensure every aircraft not only maintains safe distances, but experiences minimal delays.
“At TMAs, ATCOs are like conductors trying to manage a chaotic orchestra,” said Yi Zhang, a Senior Scientist at the A*STAR Institute for Infocomm Research (A*STAR I2R). “Besides issues of weather disruptions and human workload fatigue, TMAs involve very high volumes of aircraft taking multiple arrival and departure routes with different altitude restrictions.”
With support from the National Research Foundation and the Civil Aviation Authority of Singapore, Zhang and A*STAR I2R colleagues are developing a Practical Capacity Estimation (PCE) system: an optimisation-based decision support system to help reduce the mental load of TMA traffic management on ATCOs.
“Like a smart GPS for air traffic control, the PCE system aims to balance factors like fuel use, delays and safety when calculating the safest and most efficient routes for all planes,” Zhang explained. “The system integrates modules for AI-enhanced weather prediction, smart conflict-free routing and trajectory generation, which collectively enable TMA capacity estimation and generate feasible flight solutions.”
To refine their PCE system’s routing module, the team first created a TMA mathematical model by adapting the vehicle routing problem (VRP), a classic programming optimisation problem. Typical VRPs might depict the challenges of directing a fleet of delivery trucks with travel time and fuel efficiency in mind. For their Smart TMA Aircraft Routing Strategy (SMART) model, Zhang and colleagues added constraints unique to TMAs, such as safe time gaps between aircraft, runway schedules, anti-overtaking rules and holding patterns.
Next, the team explored multiple computational methods to find the best SMART solvers. They identified three that took different approaches: mixed-integer linear programming (MILP), multi-agent path finding (MAPF) and the evolutionary algorithm (EA).
“Think of MILP as a perfectionist planner: it tries to map out the best possible route for every plane, which can take too long for real-time situations,” said Zhang. “MAPF is a street-smart crowd navigator: it finds each plane a quick and safe path through busy skies, focusing on fast rather than fully optimal solutions. Finally, EA is an adaptive learner: it starts with many possible solutions and improves them with experience, making it useful for handling large-scale problems.”
When evaluated on a highly-congested one-hour TMA case, the three methods demonstrated improved landing times versus historical airport surveillance data. MAPF- and MILP-planned routes completed the entire landing process approximately 10 percent faster for many flights, and all three algorithms produced solutions that, at the high-level routing layer, reduced the total landing times for all aircraft by over an hour versus historical results.
Currently, the team is refining the PCE system for future inclusion in decision support tools for air navigation service providers. “We will focus on developing a fully integrated, scalable decision support tool that combines cutting-edge artificial intelligence with real-time air traffic data,” Zhang said.
The A*STAR-affiliated researchers contributing to this research are from the A*STAR Institute for Infocomm Research (A*STAR I2R).
