MOSAD: Multimodal Open-Set Anomaly Detection Dataset for Omnidirectional UAVs

MOSAD-8: Multimodal Anomaly Detection Benchmark for Edge-Deployed UAVs Dataset Overview The MOSAD-8 dataset is a stress-test benchmark designed for evaluating real-time, multimodal one-class anomaly detection frameworks on resource-constrained aerial platforms (e.g., NVIDIA Jetson Xavier NX). The dataset contains synchronized visual feature scores and 6-DOF geometric flight controller telemetry collected across 8 distinct flight sequences under severe environmental and hardware fault conditions. Total Samples / Frames: 6,000 frames (750 frames per scenario at 50 Hz cadence) Total Anomalous Frames: 2,370 frames (39.5% overall anomaly ratio) Modalities: Visual Anomaly Scores (PCA Gatekeeper + Bayesian RCAE) + 6-DOF Tracking Error Telemetry Target Platform: Omnidirectional UAVs governed by Brescianini–D'Andrea dynamics License: Creative Commons Attribution 4.0 International (CC BY 4.0) Files Included in Package scenario_1_telemetry_vision.csv : Dusk Lighting Shift (Nominal baseline) scenario_2_telemetry_vision.csv : Rapid Lens Occlusion scenario_3_telemetry_vision.csv : Motor Thrust Loss (10% single-rotor loss) scenario_4_telemetry_vision.csv : Severe Wind Gusts (Aerodynamic turbulence) scenario_5_telemetry_vision.csv : Propeller Degradation scenario_6_telemetry_vision.csv : IMU Sensor Noise Spike scenario_7_telemetry_vision.csv : Camera Gimbal Jitter scenario_8_telemetry_vision.csv : Actuator + Vision Combined Fault summary_metrics_reproduced.csv : Consolidated scenario evaluation metrics Data Schema & Column Specifications Each CSV file contains 7 synchronized telemetry and feature columns recorded at 0.02s timestamp increments (50 Hz): timestamp (float): Elapsed flight time in seconds (0.00 to 14.98 s). pca_score (float): Stage-1 Principal Component Analysis reconstruction MSE (64x64 RGB input). rcae_visual_loss (float): Stage-2 Robust Convolutional Autoencoder reconstruction error combined with Monte Carlo Dropout epistemic uncertainty variance. tracking_error_6dof_norm (float): Normalized 6-DOF tracking error derived via Mahalanobis weighting over Brescianini–D'Andrea rigid-body state dynamics. fused_anomaly_score (float): Multimodal aggregated anomaly score computed via learned weighted fusion (0.15 * PCA + 0.35 * RCAE + 0.50 * Tracking Error). decision_flag (binary): System alert output (0 = Nominal, 1 = Anomaly Detected) evaluated through a 5-frame sliding persistence window. ground_truth_label (binary): Ground-truth annotation (0 = Nominal, 1 = Anomalous Frame). Associated Publication & Citation If you use this dataset or benchmark in your research, please cite the corresponding journal publication: Omar Huseyn and Yasin Furkan Gorgulu, "Multimodal One-Class Anomaly Detection for Omnidirectional UAVs via PCA Gating and Bayesian Latent Regularization," IEEE Transactions on Circuits and Systems for Video Technology, 2026.