U2UData-2: A Scalable Swarm UAVs Autonomous Flight Dataset for Long-horizon Tasks

Tongtong Feng, Xin Wang, Feilin Han, Leping Zhang, Wenwu ZhuarXiv 2025 · arXiv · OpenReview

Abstract

Swarm UAV autonomous flight for Embodied Long-Horizon (ELH) tasks is crucial for advancing the low-altitude economy. However, existing methods focus only on specific basic tasks due to dataset limitations, failing in real-world deployment for ELH tasks. ELH tasks are not mere concatenations of basic tasks, requiring handling long-term dependencies, maintaining embodied persistent states, and adapting to dynamic goal shifts. This paper presents U2UData+, the first large-scale swarm UAV autonomous flight dataset for ELH tasks and the first scalable swarm UAV data online collection and algorithm closed-loop verification platform. The dataset is captured by 15 UAVs in autonomous collaborative flights for ELH tasks, comprising 12 scenes, 720 traces, 120 hours, 600 seconds per trajectory, 4.32M LiDAR frames, and 12.96M RGB frames. This dataset also includes brightness, temperature, humidity, smoke, and airflow values covering all flight routes. The platform supports the customization of simulators, UAVs, sensors, flight algorithms, formation modes, and ELH tasks. Through a visual control window, this platform allows users to collect customized datasets through one-click deployment online and to verify algorithms by closed-loop simulation.

Introduction

While single-UAV autonomous flight has advanced significantly, it remains vulnerable to obstacle occlusion, limited sensor range, and single-point failures. Swarm UAVs—groups of drones working in concert—address these issues through collaborative perception and task re-allocation. However, research has been constrained by datasets focusing on basic, short-duration tasks, often featuring trajectories as short as 15 seconds.

This work introduces Embodied Long-Horizon (ELH) tasks, which require swarms to maintain persistent states and adapt to shifting goals over extended periods. U2UData+ provides the necessary data and platform to support this complexity, moving beyond simple maneuvers to complex scenarios like wildlife conservation and disaster rescue.

The U2UData+ Platform

The U2UData+ platform is a highly flexible simulation environment built on Unreal Engine 5.2. It features a "Real-World Mapping Simulator" that utilizes geographical data from China’s Yunnan Province to create high-fidelity environments.

The Real-World Mapping Simulator incorporates diverse terrains, vegetation assets, and real-world meteorological data to bridge the sim-to-real gap.

The platform is designed for scalability, allowing users to customize:

  • Swarm Configuration: Support for up to 15 UAVs in collaborative flight.
  • Environmental Realism: Features 58 types of forest vegetation and 15 animal species, with weather patterns driven by data from the China Meteorological Center.
  • Operational Control: A visual control window enables "one-click" deployment and closed-loop verification, allowing researchers to test algorithms in the simulator and immediately observe results.

Dataset Characteristics

The released U2UData+ dataset represents a massive leap in scale and diversity compared to previous benchmarks, totaling 3.62 TB of data across 120 flight hours.

The Scalable Simulator interface provides multi-modal sensor data (RGB, Depth, Infrared, Segmentation) and environmental metadata for long-horizon task analysis.

Key features include:

  • Long Trajectories: Each trajectory lasts 600 seconds, providing the temporal depth needed for ELH tasks.
  • Multi-Modal Sensing: Includes 12.96M RGB/Depth frames and 4.32M LiDAR frames, supplemented by environmental sensors for airflow, temperature, and smoke.
  • Diverse Conditions: 12 distinct scenes combining 4 terrains and 7 weather conditions, such as sandstorms, snow, and heavy rain.

Benchmarking and Results

The authors evaluated 9 State-of-the-Art (SOTA) collaborative tracking models on a wildlife conservation task. The results demonstrate that collaborative perception is a force multiplier for swarm success:

  • Collaborative Advantage: Cooperative algorithms significantly outperformed single-agent baselines. The CoBEVT model achieved an Average Multi-Object Tracking Accuracy (AMOTA) of 24.63, compared to just 9.36 for the non-collaborative baseline.
  • Fusion Strategies: The benchmarks indicate that "Intermediate Fusion" (sharing processed feature maps) offers a superior balance between tracking consistency and communication bandwidth compared to late fusion. Models like V2VNet and Where2com showed high robustness in complex terrains.

Conclusion

U2UData+ addresses the critical data gap for swarm UAVs performing long-horizon tasks. By providing a high-fidelity, scalable simulation platform and a massive multi-modal dataset, it enables the development of more robust collaborative algorithms. Future work will focus on expanding the dataset to more ELH tasks like logistics and patrol, and validating these models in real-world forest environments.

Summary updated

Abstract

Swarm UAV autonomous flight for Embodied Long-Horizon (ELH) tasks is crucial for advancing the low-altitude economy. However, existing methods focus only on specific basic tasks due to dataset limitations, failing in real-world deployment for ELH tasks. ELH tasks are not mere concatenations of basic tasks, requiring handling long-term dependencies, maintaining embodied persistent states, and adapting to dynamic goal shifts. This paper presents U2UData+, the first large-scale swarm UAV autonomous flight dataset for ELH tasks and the first scalable swarm UAV data online collection and algorithm closed-loop verification platform. The dataset is captured by 15 UAVs in autonomous collaborative flights for ELH tasks, comprising 12 scenes, 720 traces, 120 hours, 600 seconds per trajectory, 4.32M LiDAR frames, and 12.96M RGB frames. This dataset also includes brightness, temperature, humidity, smoke, and airflow values covering all flight routes. The platform supports the customization of simulators, UAVs, sensors, flight algorithms, formation modes, and ELH tasks. Through a visual control window, this platform allows users to collect customized datasets through one-click deployment online and to verify algorithms by closed-loop simulation. U2UData+ also introduces an ELH task for wildlife conservation and provides comprehensive benchmarks with 9 SOTA models. U2UData+ can be found at https://fengtt42.github.io/U2UData-2/.
BibTeX
@misc{feng2025u,
  author = {Tongtong Feng and Xin Wang and Feilin Han and Leping Zhang and Wenwu Zhu},
  title = {{U2UData-2: A Scalable Swarm UAVs Autonomous Flight Dataset for Long-horizon Tasks}},
  year = {2025},
  eprint = {2509.00055},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2509.00055},
}
Figure