Research map / 3D vision / 3D generation

GeoWind2Plan: Mission-Time 3D Urban Wind Prediction for Energy-Efficient UAV Planning

NeurIPS 2026 · 3D vision › 3D generation · approach: 2D-to-3D pretrained representations · 0 citations (OpenAlex)

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes GeoWind2Plan, which predicts corridor-localized 3D urban wind with a geometry-conditioned FNO in a wind-aligned frame and plans energy-efficient UAV trajectories with RRT-initialized continuous optimization, evaluated under CFD wind.

Problem

Urban UAV energy-efficient planning requires building-resolved 3D wind, but CFD is too slow (~8 hours per 1.2 km case) and tied to fixed inflow boundary conditions for mission-time use.

Approach

Combines yaw-covariance (rotating city geometry into a reference-wind frame), mean-flow speed rescaling, and a patch-based geometry-conditioned 3D FNO that predicts only mission-relevant corridor patches, feeding a continuous trajectory and speed optimizer that is judged by CFD-evaluated energy.

Key results

• Under CFD evaluation, Ours reduces energy vs wind-agnostic by 6.9%, 12.7%, and 4.5% in tailwind, headwind, and crosswind missions, recovering 87.9%, 85.7%, and 75.0% of CFD-reference savings.
• Corridor-localized wind inference takes about 3 seconds versus roughly 8 hours for CFD on a 1.2 km case.
• At 8 m/s headwind on block A, energy overhead vs ground truth is 7.6% for Ours versus 41.0% for wind-agnostic planning.

Surprising findings

• Wind-aware planners fly higher and faster in tailwinds and lower and slower in headwinds, and mean drag can be higher in tailwinds while total energy still drops due to shorter mission time.
• Corridor width reduction from full field to 20% leaves CFD-evaluated energy nearly unchanged, suggesting energy-optimal routes stay near the direct start-goal line.

Limitations

• Uses time-averaged velocity fields; no turbulence, gusts, or turbulence-related safety risk modeled.
• Assumes background wind direction and speed remain fixed during a mission.
• Narrow corridors (10%) cause planning failures when tall buildings block all feasible paths.

Experimental setup

Datasets: CityFFD-generated urban CFD wind fields (~20 1.2 km cases plus one 3 km case)

Benchmarks: urban blocks A-E held-out CFD domains

Metrics: unit-distance energy consumption (Wh/km) inference time fraction of CFD-reference savings recovered

Baselines: Wind-agnostic planning Vertical profile wind Ground truth CFD planning

Builds on: Fourier neural operator GINO CityFFD RRT multi-directional distance features

Closest papers by meaning

• DrivAerStar: An Industrial-Grade CFD Dataset for Vehicle Aerodynamic Optimization (NeurIPS 2025 · similarity 0.81)
• AeroGTO: An Efficient Graph-Transformer Operator for Learning Large-Scale Aerodynamics of 3D Vehicle Geometries (AAAI 2025 · similarity 0.81)
• U2UData+: A Scalable Swarm UAVs Autonomous Flight Dataset for Embodied Long-horizon Tasks (AAAI 2026 · similarity 0.81)
• Weather-Robust Cross-View Geo-Localization via Prototype-Based Semantic Part Discovery (NeurIPS 2026 · similarity 0.80)
• 3DID: Direct 3D Inverse Design for Aerodynamics with Physics-Aware Optimization (NeurIPS 2025 · similarity 0.80)
• SatNav: A Scalable Benchmark for Long-Horizon UAV Vision-Language Navigation from Satellite Imagery (NeurIPS 2026 · similarity 0.80)