DroneLog: A Flight Telemetry Benchmark Dataset for UAV Airframe Identification and Payload State Recognitio

DroneLog is a dual-task UAV flight telemetry benchmark designed for airframe identification and payload state recognition. The benchmark contains 50 real-world flight sessions collected from five quadrotor platforms with frame sizes of 250, 450, 500, 650, and 680 mm, all equipped with PX4-based Pixhawk flight controllers. The benchmark is organized into two sub-datasets: Frame Size Sub-dataset:Contains 15 flight sessions, comprising three flights for each of the five frame-size classes. The recordings provide a total flight duration of 28 minutes and 11 seconds and approximately 1.3 million rows of telemetry data in CSV format. Payload State Sub-dataset:Contains 35 flight sessions collected using the Holybro X500 V2 (500 mm) and Holybro X650 (650 mm) platforms across five operational scenarios: no payload, full-flight payload transport, pickup-and-carry, pickup-and-drop, and carry-and-drop. Payload configurations include 0.5, 0.75, 1, 1.5, and 2 kg conditions. The recordings provide a total flight duration of 1 hour and 49 minutes with approximately 4.2 million telemetry rows. The benchmark is distributed primarily as pre-extracted CSV telemetry files, where each PX4 telemetry topic is stored separately for every flight session. Original PX4 ULog binaries are additionally provided as supplementary material. For payload scenarios involving mid-flight state transitions, pickup and drop event timestamps are manually annotated. This benchmark accompanies the paper entitled “DroneLog: A Flight Telemetry Benchmark Dataset for UAV Airframe Identification and Payload State Recognition”, submitted to The International Journal of Robotics Research (2026).