Context-Aware Remote ID Anomaly Detection for UAS Activity Monitoring Using Multi-Agent LLM Reasoning
Year: 2026
Type: conference-paper
Abstract: This paper presents a context-aware anomaly detection framework for identifying potentially unsafe or suspicious unmanned aircraft system (UAS) activity using broadcast Remote ID (RID) messages, with a focus on operations near critical infrastructure such as airports. The RID provides standardized telemetry and identification, interpreting observed behavior in real-world environments requires more than threshold-based kinematic checks; risk depends on surrounding context (e.g., proximity to sensitive assets, runway protection surfaces, time-of-day patterns, and weather conditions). To address this gap, the paper proposes a hybrid analytics and reasoning architecture that combines deterministic trajectory analysis with large language model (LLM) agents to produce interpretable, operationally actionable assessments suitable for counter-UAS (C-UAS) monitoring and incident triage. The system uses ASTM-compliant RID data captured by a receiver network, normalizes multi-message RID streams into coherent tracks, and computes behavior features, and a probabilistic indicator will be calculated. These features are fused with contextual layers comprising geofenced airport surfaces and asset buffers, criticality-weighted infrastructure maps, and weather-derived explainability factors (e.g., wind-driven drift likelihood) to generate risk-weighted anomaly scores that adapt to location, time, and operating conditions. On top of the quantitative anomaly engine, we introduce a multi-agent LLM layer that performs structured reasoning and reporting. To support development and evaluation, we also implement a RID simulator that generates ASTM-format message streams for scenario-driven testing, including nonsuspicious baselines and airport-adjacent anomalies such as perimeter loitering near facilities, sustained heading alignment (e.g. EO sensor) toward sensitive zones, repeated corridor incursions, multi-UAS coordination, and reception impairments (dropouts/noise) representative of cluttered environments. As a proof-of-concept demonstration, we evaluate the framework on a coordinated dual-UAS incursion scenario at Grand Forks International Airport. The proposed approach enables context-sensitive detection and human-interpretable explanations of the practicality of RID-based monitoring as an analytics component in C-UAS workflows.
Institution: University of North Dakota
Language: English
Subfield: Aerospace Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7165753181
