TL;DR
Fair-CoPlan is a new semi-distributed flight planning framework for Urban Air Mobility (UAM) that balances operator flexibility with airspace safety. By using a three-step negotiated process modeled via Mixed Integer Linear Programs (MILPs), it allows drone operators to design their own routes while ensuring the central authority deconflicts overlapping paths in a proportionally fair manner.
What Problem Does This Paper Address?
As Urban Air Mobility (UAM) matures, the skies above cities will soon be shared by autonomous Uncrewed Aerial Systems (UAS) managed by different commercial operators. Traditional Air Flow Traffic Management (AFTM) relies on a centralized authority to dictate rigid flight paths. This approach strips operators of flexibility and can result in highly unfair outcomes—for instance, forcing a small operator's short delivery flight into a massive detour to clear the way for a larger carrier.
To encourage widespread participation in UAM, we need a system that respects operator autonomy and distributes the "cost of deconfliction" (delays and detours) equitably, all without compromising airspace safety.
Methods
The authors introduce Fair-CoPlan, a semi-distributed, three-step negotiated framework implemented using Mixed Integer Linear Programming (MILP):
- PSU Constraints (Centralized): The Provider of Service to the UAM (PSU) analyzes high-congestion zones—specifically vertiports and their immediate surroundings. It then provides operators with a flexible window of feasible departure/arrival times and approved entry/exit sectors.
- Operator Planning (Distributed): Within these constraints, individual operators independently calculate their optimal trajectories to minimize their own Total Delay Cost (TDC).
- Fair Deconfliction (Centralized): The PSU aggregates the proposed plans and resolves en-route conflicts. Crucially, the objective function minimizes both the actual delay and a novel proportional fairness term ($F$).
Instead of just equalizing absolute delay, Fair-CoPlan defines fairness based on the relative change in path length ($\frac{\text{Actual Length}}{\text{Proposed Length}}$). This ensures that shorter flights aren't hit with proportionally devastating detours. The authors formulate $F$ as a piecewise linear function (the difference between the maximum and minimum relative path changes) to keep the optimization computationally tractable.
Datasets
Because real-world UAM data does not yet exist at scale, the authors evaluated their framework using a simulated environment:
- Airspace: A $3600\text{ km}^2$ region discretized into a $15 \times 15$ grid, modeling a metro area like Marseille or Istanbul.
- Infrastructure: 12 vertiports (4 high-capacity hubs, 8 smaller "vertistops").
- Traffic Demand: Scenarios testing 25, 36, and 50 flights per hour per hub vertiport across 10 multi-hour simulated operating days.
Key Insights and Ideas
- The Power of Negotiation: Allowing operators to propose their own paths after the central authority constrains vertiport access yields significantly lower overall delay costs than traditional, fully centralized traffic management systems.
- Proportional vs. Absolute Fairness: Treating a 5-minute flight and a 50-minute flight identically when handing out 10-minute delays is fundamentally unfair. Proportional fairness preserves the viability of short-range UAM missions.
- Tractable Math Matters: While calculating fairness using variance (a quadratic function) is intuitive, it drastically slows down optimization. Switching to a piecewise linear max-min difference formulation keeps solve times practical for real-time deployment.
Key Results
- Lower Delays: Both CoPlan (the baseline three-step framework) and Fair-CoPlan achieved significantly lower average Total Delay Costs per flight compared to classical centralized AFTM formulations (TFMP), which tend to hold aircraft in sectors rather than allowing smart diversions.
- Quantifiable Fairness: Fair-CoPlan improved path equity in up to 93% of simulated days for medium-demand scenarios (36 flights/hr) compared to the non-fair baseline.
- Operational Efficiency: The entire three-step pipeline runs in under 5 minutes. Vertiport allocation and operator path planning take less than 45 seconds, while the final en-route deconfliction step finishes in under 3 minutes, making it highly suitable for rolling 5-minute UAM planning windows.
Limitations and Future Work
A key limitation is that Fair-CoPlan evaluates fairness myopically—only considering flights within the current 5-minute planning window. A fair choice made now might inadvertently block the airspace and cause massive, unfair delays for flights arriving in the next window.
The authors note that future iterations will focus on incorporating predictive models for future flight demands, scaling the framework to handle higher time granularity, and introducing guardrails against "adversarial planning" (where a single operator floods the system with requests to manipulate the fairness metric).