{"success":true,"data":{"author":"Aykut Kabaoglu and Sanem Sariel","bucketAccessType":"public","bucketName":"aykutkabaoglu/uav-flight-anomaly-dataset","bucketPrefix":"","category":"Robotics","cloudProvider":"huggingface","createdAt":"2026-11-21T22:15:33.086Z","description":"UAV-SEAD is a large real-world flight dataset for fault detection and identification (FDI) in unmanned aerial vehicles, focused on failures of the onboard state estimator, the part of the autopilot that fuses sensors into position and attitude. It contains 1,396 categorized flight logs, more than 52 hours of flight, from a fleet of custom PX4-based multirotor drones with different airframes, motor sizes, weights and sensor suites, flown in indoor hangars, warehouses and outdoors; most flights have external position data. Anomalies are point anomalies (sensor spikes and outliers), contextual anomalies (readings anomalous only for the flight state) and collective anomalies (sustained drift indicating gradual failure), labeled by physically constrained expert evaluation that compared autopilot estimates with observed outcomes. The data are raw PX4 ULog (.ulg) files under ulg_files/ with CSV summaries, PDF plots and a mapping.json, about 4,240 files and 30 GB (measured 2026-11-21). Convert ULogs to CSV and annotate with the authors' ulog_annotation_tool or pyulog. Use it for multivariate time-series anomaly detection and predictive maintenance of drones. Released under CC BY 4.0; cite Kabaoglu and Sariel, UAV-SEAD.","displayTitle":"UAV-SEAD: State Estimation Anomaly Dataset of 1,396 Real PX4 Multirotor Flight Logs for Fault Detection","huggingFaceUrl":"https://huggingface.co/datasets/aykutkabaoglu/uav-flight-anomaly-dataset","id":"a04e7123-929e-5985-b8a7-b590b90649b0","license":"CC-BY-4.0","licenseEvidenceUrl":"https://huggingface.co/datasets/aykutkabaoglu/uav-flight-anomaly-dataset","objectCountEstimate":4240,"probeStatus":"ok","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"aykutkabaoglu/uav-flight-anomaly-dataset\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"aykutkabaoglu/uav-flight-anomaly-dataset\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=aykutkabaoglu, license:cc-by-4.0, size_categories:n<1K, downloads=2862"],"format":"mixed","generatedAt":"2026-11-21T22:31:30.788Z","layout":{"example":"ulg_files/2018-06-04/18_01_59.ulg.mdat.csv","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"name,value,description\nfilter_faults_max,0,Largest recorded value of the filter internal fault bitmask. Should always be zero.\nflow_sensor_status,Pass,Optical Flow sensor check summary. A Fail result indicates a significant error that caused a significant reduction in vehicle navigation performance was detected. A Warning result indicates that error levels higher than normal were detected but these errors did not significantly impact navigation performance. A Pass result indicates that no amonalies were det","encoding":"utf8","key":"ulg_files/2018-06-04/18_01_59.ulg.mdat.csv","truncatedAt":512}],"sampleObjects":["mapping.json","ulg_files/00_02_49.ulg","ulg_files/00_14_08.ulg","ulg_files/2018-05-24/18_57_50.ulg","ulg_files/2018-05-24/18_59_29.ulg"],"sidecars":[{"bytes":"---\nlanguage:\n- en\nlicense: cc-by-4.0\ntags:\n- state-estimation-anomaly\n- uav-anomaly\n- anomaly-detection\n- fault-detection\n- multivariate-time-series\n- uav-anomaly-dataset\n- px4-anomaly-dataset\n- uav-dataset\n- px4-dataset\n---\n\n\n# DATASET CARD for UAV-SEAD\n\n## Citation\n\n**BibTeX:**\n```bibtex\n@article{kabaoglu2024uavsead,\n  title={UAV-SEAD: State Estimation Anomaly Dataset for UAVs},\n  author={Kabaoglu, Aykut and Sariel, Sanem},\n  journal={arXiv preprint},\n  year={2026}\n}\n```\n\n## DATASET INFORMATION\n\n### Dataset Description\nUAV-SEAD (State Estimation Anomaly Dataset) is a comprehensive, large-scale, real-world flight dataset designed to advance research in Fault Detection and Identification (FDI) for Unmanned Aerial Vehicles. It contains 1,396 categorized real-world flight logs totaling over 52.4 hours of flight data. Unlike existing benchmarks, it avoids synthetic fault injection, providing naturally occurring anomalies across diverse environmental and hardware configurations.\n\n- **Curated by:** Aykut Kabaoglu\n- **License:** Creative Commons Attribution 4.0 International (CC BY 4.0)\n\n### Dataset Sources\n- **Paper:** \"UAV-SEAD: State Estimation Anomaly Dataset for UAVs\"\n\n### Toolbox for Dataset Preprocessing and Annotation\n- https://github.com/aykutkabaoglu/ulog_annotation_tool\n- Use readme in the `ulog_annotation_tool` repository to start using dataset. The first step is converting raw flight logs into csv files. Then the flight logs can be used for plotting and labeling.\n\n## DATASET CREATION AND USAGE\n\n### Dataset Structure\n\nThe dataset is structured as a collection of multivariate time-series sequences. \n- **Point Anomalies:** Spikes and outliers in sensor readings.\n- **Contextual Anomalies:** Readings that are only anomalous relative to the flight state (e.g., zero altitude during thrust).\n- **Collective Anomalies:** Sustained drifts or patterns indicating gradual system failure.\n\n### Use Cases\n- Benchmarking multivariate time-series anomaly detection algorithms.\n- Developing predictive maintenance and fault identification systems for UAVs.\n\n#### Data Collection and Processing\nData was harvested from a fleet of custom-built, PX4-based UAVs. \n- **Hardware:** Multiple airframes with variations in motor sizes, weights, and sensor suites. Check paper for hardware details.\n- **Environments:** Flight missions conducted in indoor hangars, warehouses, and open outdoor spaces.\n\n#### Annotation process\nAnnotation was performed via **Physically-Constrained Expert Evaluation**. Human experts identified state estimation divergences by comparing autopilot estimates with real-world observed outcomes. Labels were verified against physical flight dynamics to ensure the statistical anomalies correlate with actual physical events.\n\n### Bias, Risks, and Limitations of Data Collection\n- **Platform Specificity:** Limited to multi-rotor dynamics. Limited outdoor flights with GPS availability.\n\n\n## DATASET STATISTICS\n\n### Annotated Data Flight Source Statistics\nAnnotated Number of Flight: 1396\n\n| Type | Time (hh:mm:ss)} |\n|--------|----:|\n| Total flight duration | 52:20:05 |\n| External position data duration | 51:28:34 |\n| Global position data duration | 04:35:24 |\n| Distance sensor data duration | 51:35:05 |\n| Barometer altitude data duration | 26:23:45 |\n\n### Annotated Data Class Statistics\n\n| Class Labeling | # Flights | Total Flight Duration (hh:mm:ss)} | Anomaly Duration (hh:mm:ss) |\n|--------|----|---------:|----------:|\n| Normal | 900 | 38:24:19 |  -- | -- | -- |\n| Mechanical and Electrical | 47 | 02:12:38 | 00:12:43 |\n| External Position | 197 | 05:12:25 | 01:17:02 |\n| Global Position | 41 | 01:40:42 | 00:36:23 |\n| Altitude | 78 | 02:41:21 | 00:25:57 |\n| Uncategorized | 141 | 03:13:09 | -- |\n\n\n### RAW Data Statistics\nNumber of Flights: 3196\n\n### RAW Data, Data Fields\nThe dataset follows a **Superset Schema** of 1897 unique columns. Fields are grouped into 84 unique topics by PX4 UORB topic prefixes.\n\n81 PX4 Topics out of 84 and their fields that has data at least a single flight log in the dataset are listed as below:\n\n| # | Topic (Su","encoding":"utf8","key":"README.md","name":"README.md","size":30655,"truncated":true}],"sourceObjectKey":"ulg_files/2018-06-04/18_01_59.ulg.mdat.csv","starterCell":"import pandas as pd\ndf = pd.read_csv(\"hf://datasets/aykutkabaoglu/uav-flight-anomaly-dataset/ulg_files/2018-06-04/18_01_59.ulg.mdat.csv\")\ndf.head()","starterSamplePath":"mapping.json","totalObjects":4240},"summary":"UAV-SEAD is a large real-world flight dataset for fault detection and identification (FDI) in unmanned aerial vehicles, focused on failures of the onboard state estimator, the…","title":"UAV-SEAD: State Estimation Anomaly Dataset of 1,396 Real PX4 Multirotor Flight Logs for Fault Detection","updatedAt":"2026-11-22T00:55:10.581Z","visibility":"public","websiteUrl":"https://github.com/aykutkabaoglu/ulog_annotation_tool"}}