Uncovering drone intentions using control physics informed machine learning is an open-access rapport by Adolfo Perrusquía, Weisi Guo, Benjamin Fraser, and Zhuangkun Wei.
Unmanned Autonomous Vehicles (UAVs), commonly known as drones, are increasingly utilized across various sectors. However, uncooperative drones do not disclose their identity or flight plans and pose significant risks to critical infrastructure. Assessing a drone’s intention is crucial for risk management and executing appropriate contre-mesures. While intentions are often intangible and unobservable, they can be inferred through tangible intention classes.
Traditional methods that rely solely on observational data to infer these intention classes are inherently unreliable due to biases in observation and learning. The authors have developed a control-physics-informed machine learning (CPhy-ML) model to address this. This model robustly infers intention classes by combining the representational power of deep learning with the conservation laws of aerospace models, thus reducing bias and instability.
The CPhy-ML model achieves a 48.28% improvement in performance over traditional trajectory prediction methods, marking a significant advancement in the reliable inference of drone intentions.
Uncovering drone intentions using control physics informed machine learning contains the following major sections:
- Introduction
- Résultats
- Discussion
- Méthodes
- Disponibilité des données
- Code availability
Post Image- Trajectory Intention Regression example results (Post Image Credit: Authors)
