Research map / Multi-agent systems and games / LLM agents

U2UData+: A Scalable Swarm UAVs Autonomous Flight Dataset for Embodied Long-horizon Tasks

AAAI 2026 · Multi-agent systems and games › LLM agents · approach: Software engineering agent benchmarks · 7 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.

Introduces U2UData+, a large-scale swarm UAV flight dataset for embodied long-horizon wildlife-conservation tasks plus an online collection and closed-loop simulation platform, benchmarking nine collaborative tracking methods.

Problem

Existing swarm UAV datasets cover only basic perception/tracking tasks with short, fixed trajectories and cannot be expanded, which blocks research on embodied long-horizon (ELH) tasks for real-world deployment.

Approach

A scalable UE5.2-based real-world-mapped simulator and visual control window that lets users one-click collect customized multi-sensor swarm UAV data and run closed-loop algorithm verification, released together with a 15-UAV, 600-second-per-trajectory ELH dataset.

Key results

• Where2com and CoBEVT reach the top MOTA of 50.82 and 51.18 respectively among fusion methods, versus 23.47 MOTA for No Fusion with AMOTA 9.36.
• V2VNet attains the best sAMOTA of 57.82 and MT of 35.68.
• Intermediate fusion methods improve AMOTA by at least 35.97% over No Fusion, per the paper's text.

Surprising findings

• Early Fusion underperforms Late Fusion on sAMOTA (43.80 vs 47.96) despite higher AMOTA, showing metric disagreement.

Limitations

• Only one ELH task (wildlife conservation) is actually collected; other ELH tasks are presented as platform capabilities without data.
• Benchmark covers only collaborative tracking, not full ELH task execution.
• Simulated data with a domain gap to real-world deployment remains unvalidated in the text.

Experimental setup

Datasets: U2UData+

Benchmarks: U2UData+ swarm UAV collaborative tracking benchmark

Metrics: AMOTA AMOTP sAMOTA MOTA MT ML

Baselines: No Fusion Late Fusion Early Fusion When2Com DiscoNet V2VNet V2X-ViT CoBEVT Where2com

Builds on: U2USim U2UData AB3DMOT PointPillar

Closest papers by meaning

• UAV-Flow Colosseo: A Real-World Benchmark for Flying-on-a-Word UAV Imitation Learning (NeurIPS 2025 · similarity 0.84)
• Towards Realistic UAV Vision-Language Navigation: Platform, Benchmark, and Methodology (ICLR 2025 · similarity 0.83)
• AGC-Drive: A Large-Scale Dataset for Real-World Aerial-Ground Collaboration in Driving Scenarios (NeurIPS 2025 · similarity 0.83)
• SatNav: A Scalable Benchmark for Long-Horizon UAV Vision-Language Navigation from Satellite Imagery (NeurIPS 2026 · similarity 0.82)
• UAV3D: A Large-scale 3D Perception Benchmark for Unmanned Aerial Vehicles (NeurIPS 2024 · similarity 0.82)
• Is your VLM Sky-Ready? A Comprehensive Spatial Intelligence Benchmark for UAV Navigation (CVPR 2026 · similarity 0.82)