In urban low-altitude flight, buildings reshape ambient wind into spatially varying 3D flow, making unmanned aerial vehicle (UAV) energy depend on local wind exposure as well as path length. However, building-resolved wind information is rarely available when a mission must be planned. Computational fluid dynamics (CFD) can produce high-fidelity urban flow fields, but each simulation is tied to a fixed inflow boundary condition and can take hours to days, which is incompatible with urban UAV missions that typically last minutes to tens of minutes. We present GeoWind2Plan, a geometry-to-wind-to-planning framework for mission-time 3D urban wind prediction and energy-efficient UAV planning. Given only a background wind vector, 3D building geometry, and a start-goal pair, GeoWind2Plan transforms the building geometry into a reference-wind frame, predicts mission-relevant 3D wind patches with a localized geometry-conditioned neural operator, stitches them into a queryable local wind field, and optimizes a feasible 3D path and speed profile using a physically grounded UAV energy model. Rather than pursuing CFD-perfect reconstruction, GeoWind2Plan targets decision-useful wind prediction: trajectories are planned with predicted wind and evaluated under high-fidelity CFD wind. Across held-out urban domains, wind speeds, and mission wind-angle regimes, GeoWind2Plan performs corridor-localized wind inference in about 3 seconds, compared with roughly 8 hours for CFD. Under CFD evaluation, trajectories planned with GeoWind2Plan reduce energy by 6.9%, 12.7%, and 4.5% in tailwind, headwind, and crosswind missions relative to wind-agnostic planning, recovering 87.9%, 85.7%, and 75.0% of CFD-reference savings. These results show that fast, corridor-localized 3D urban wind prediction can make wind-aware UAV energy planning practical at mission time.
Figures & tables
Figure 1 : Overview of the GeoWind2Plan framework. Given a start-goal pair, 3D building geometry, and a background wind vector, GeoWind2Plan extracts a mission-relevant corridor, rotates the local geometry into a reference-wind frame, predicts corridor-relevant wind patches with a localized geometry-conditioned neural operator, stitches them into a queryable local 3D wind field, rescales the field from the reference training speed to the given background wind speed, and rotates the predicted wind field back to the city frame. The resulting local wind field is used by an RRT-initialized continuous optimizer to produce an energy-efficient 3D UAV path and speed profile.
Urban block
U(m/s)
Method
Unit-distance energy consumption (Wh/km)
All
Tailwind
Headwind
Crosswind
A
2
Ground truth
12.4 ± 0.8
11.3 ± 0.4
13.2 ± 0.3
12.5 ± 0.6
Wind-agnostic
12.7 ± 0.8 (2.4% ↑ )
11.7 ± 0.3 (3.1% ↑ )
13.7 ± 0.4 (3.8% ↑ )
12.7 ± 0.6 (1.5% ↑ )
Profile wind
12.6 ± 0.8 (1.3% ↑ )
11.4 ± 0.4 (1.2% ↑ )
13.4 ± 0.4 (2.1% ↑ )
12.6 ± 0.6 (1.0% ↑ )
Ours
12.5 ± 0.8 (0.4% ↑ )
11.3 ± 0.4 (0.2% ↑ )
13.3 ± 0.3 (0.7% ↑ )
12.5 ± 0.6 (0.3% ↑ )
4
Ground truth
12.3 ± 1.4
10.2 ± 0.6
13.7 ± 0.5
12.4 ± 1.0
Table 1 : Quantitative evaluation of UAV trajectory-planning efficiency across urban blocks A, B, and E at wind speed U=4m/s and relative wind-angle regimes (see Appendix Table 4 for C and D). The table reports unit-distance energy consumption (Wh/km) as mean ± standard deviation, with extra energy measured relative to Ground truth reference. The best results are highlighted in bold.
Figure 2 : Qualitative trajectory-planning comparison in urban block D. Panels (a)–(d) show four randomly selected start–goal missions with different relative wind angles. Each pair compares the trajectory planned with the offline CFD reference wind field ( Ground truth ) and the trajectory planned with GeoWind2Plan’s predicted 3D wind field ( Ours ). Wind-speed slices and horizontal vectors at 75m are shown as context, with optimized UAV trajectories overlaid.
Figure 3 : Representative trajectories and trajectory statistics under different relative wind angles.
Method
Corridor Width
Energy (Wh/km)
Ground truth
Full field
12.3±1.4
Ours
Full field
12.5±1.5
Ours
50%
12.4±1.5
Ours
20%
12.4±1.6
Ours
10%
12.4±1.9
Wind-agnostic
Full field
13.3±1.8
Table 2 : UAV unit-distance energy consumption under varying corridor width constraints on urban block A, averaged over all relative wind-angle regimes.
Appendix figures & tables8 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 4 : Additional task trajectory visualization over urban blocks A–E. Four tasks are randomly selected from each urban block, demonstrating generalizability across different urban environments and start–goal settings.
Figure 5 : Additional trajectory-planning comparisons across 40 randomly sampled missions in urban block D. Each panel overlays trajectories planned with four different planning-time wind inputs: offline CFD reference wind ( GT ), GeoWind2Plan-predicted local 3D wind ( Ours ), height-only profile wind ( Profile ), and zero wind ( Agnostic ).
Mean Altitude (m)
Mean Speed (m/s)
Urban block
Height (m)
Wind speed
Condition
Ground truth
Ours
Profile wind
Wind agnostic
Ground truth
Ours
Profile wind
Wind agnostic
A
50
4
Overall
71.4
71.4
63.1
66.8
13.91
13.92
13.82
13.53
Tailwind
100.7
102.3
108.2
65.5
15.71
15.89
15.50
13.53
Headwind
53.6
53.5
34.7
69.1
12.90
12.80
12.88
13.54
Crosswind
69.4
69.0
60.6
66.1
13.75
13.74
13.67
13.53
75
2
Overall
86.7
86.1
75.0
87.6
13.65
13.68
13.66
13.54
Appendix
Table 3 : Additional statistics of flight characteristics under different wind conditions, heights, and wind speeds, including altitude and speed.
Urban block
U(m/s)
Method
Unit-distance energy consumption (Wh/km)
All
Tailwind
Headwind
Crosswind
A
2
Ground truth
12.4 ± 0.8
11.3 ± 0.4
13.2 ± 0.3
12.5 ± 0.6
Wind-agnostic
12.7 ± 0.8 (2.4% ↑ )
11.7 ± 0.3 (3.1% ↑ )
13.7 ± 0.4 (3.8% ↑ )
12.7 ± 0.6 (1.5% ↑ )
Profile wind
12.6 ± 0.8 (1.3% ↑ )
11.4 ± 0.4 (1.2% ↑ )
13.4 ± 0.4 (2.1% ↑ )
12.6 ± 0.6 (1.0% ↑ )
Ours
12.5 ± 0.8 (0.4% ↑ )
11.3 ± 0.4 (0.2% ↑ )
13.3 ± 0.3 (0.7% ↑ )
12.5 ± 0.6 (0.3% ↑ )
4
Ground truth
12.3 ± 1.4
10.2 ± 0.6
13.7 ± 0.5
12.4 ± 1.0
Appendix
Table 4 : Quantitative evaluation of UAV trajectory-planning efficiency across five urban blocks (A–E), wind speeds U∈{2,4,8}m/s , and relative wind-angle regimes. The table reports unit-distance energy consumption (Wh/km) as mean ± standard deviation, with extra energy measured relative to Ground truth reference. The best results are highlighted in bold.
Figure 6 : 3D building geometry visualizations of the five randomly selected evaluation urban blocks used in the experiments: four standard 1.2km×1.2km blocks (A–D) and one larger 3km×3km block (E).
Pair
Ua
Ub
Uref
ϵvec
αmean
2 vs. 4
4
2
4
0.0669
0.1115rad(6.39∘)
8 vs. 4
4
8
4
0.1124
0.1365rad(7.82∘)
Appendix
Table 5 : Metrics for normalized CFD wind-field agreement under speed rescaling. The reference speed is Uref=4 .
Figure 7 : Comparison of raw and normalized CFD wind velocity components across varying inlet speeds. The raw spatial flow patterns (u,v,w) are highly similar and differ primarily in absolute scale, while normalizing by the corresponding inlet speeds produces aligned components (u~,v~,w~) across different speeds.
Symbol
Meaning
Urban domain and geometry
Ω⊂R3
3D urban computational domain.
Ωfly
Feasible UAV flight envelope, restricted to free space and the allowed altitude band.
Ωtask
Task-relevant region used for local wind inference and planning.
x , y
Spatial locations in the original city frame and the reference-wind frame, respectively.
G
Building occupancy field; G(x)=1 denotes building space and G(x)=0 denotes free space.
Advanced air mobility operations hold the potential to enhance and expand regional transportation of both people and goods in populated areas. However, hazardous flight conditions arising from interactions between wind and the built environment remain a significant challenge for aerial vehicles in urban settings. This work proposes a novel framework towards safe flight planning of aerial vehicles in windy urban environments. A learning-based surrogate model is trained to rapidly predict flow fields from readily available information such as building geometry and incident wind. This surrogate prediction is used to calculate a volumetric flight challenge scalar field based on critical flow parameters and proximity to structures. A safe, flow-informed flight trajectory is then identified through a cost-minimizing pathfinder. The complete system is demonstrated experimentally through flight tests of a micro aerial vehicle through a model urban geometry placed in a large fan-array wind tunnel. Comparing this approach to trajectories generated without knowledge of the wind field, we find the flow-informed approach reduces undesired vehicle displacement and improves flight stability. This work is among the first practical demonstrations of safe, wind-aware methodologies for advanced air mobility in urban environments.
Peter I. Renn, Alejandro A. Stefan-Zavala, Julian Humml +8
1: California Institute of Technology, Pasadena, CA, USA · 2: Technology Innovation Institute, Abu Dhabi, UAE
This work presents a motion planning framework for UAV navigation in non-convex urban air corridors. The planner is based on a mixed-integer tracking model predictive control formulation that enforces corridor feasibility and dynamic consistency within a single optimization problem. To guarantee convergence to the target and mitigate the occurrence of local minima induced by non-convex geometry, a shortest-path-based offset cost with feasibility constraints is embedded directly into the planning problem. Numerical simulations show that the proposed formulation generates dynamically valid trajectories that satisfy the corridor constraints and converge to the target without relying on external global planning stages.
Henrique Silva, Marcelo A. Santos, Guilherme V. Raffo
Graduate Program in Electrical Engineering, Universidade Federal de Minas Gerais, Belo Horizonte, MG, Brazil · Department of Management, Information and Production Engineering, University of Bergamo, Dalmine, BG, Italy · Department of Electronic Engineering, Universidade Federal de Minas Gerais, Belo Horizonte, MG, Brazil
Safe autonomous Uncrewed Aerial Vehicle (UAV) navigation in urban environments requires real-time path planning that avoids obstacles. MaxConvNet is a potential-field planner that leverages properties of Maxwell's equations to generate a path to the goal without local minima. We extend the 2D MaxConvNet magnetic field planner to 3D, using a convolutional autoencoder to predict obstacle-aware potential fields from LiDAR-derived 101^3 voxel grids. Evaluation across 100 randomized closed-loop trials in two distinct Cosys-AirSim urban environments, a dense night-time cityscape and a suburban district shows a 100% path planning success rate on both maps without retraining. In offline path planning, 3DMaxConvNet produces path lengths comparable to A* on unseen maps while reducing runtime from 0.155--0.17s to 0.087--0.089s, or about 1.7--1.95 times faster than A*. Against RRT*(3k), 3DMaxConvNet achieves similar path quality while reducing planning runtime from 17.2--17.5s to about 0.09s, which is roughly 193--201 times faster than RRT*(3k).
Haechan Mark Bong, Giovanni Beltrame
Department of Computer Engineering and Software Engineering, Polytechnique Montréal, QC., Canada. · MILA, QC., Canada.