Abstract
Passive tracking using Angle of Arrival (AOA) sensors is a critical capability for stealthy surveillance and electronic reconnaissance. However, coordinating a team of Unmanned Aerial Vehicles (UAVs) to track multiple targets in 3D space involves significant challenges, including measurement asynchronicity, complex physical constraints, and high computational costs. This paper introduces a fast, collaborative trajectory optimization framework that combines a novel asynchronous tracking model with a "Comprehensive Solution Method" (CSM). Experimental results show that the CSM achieves tracking accuracy comparable to global search algorithms while reducing computational time by over three orders of magnitude, making real-time 3D aerial tracking feasible.
Introduction
Passive tracking—locating targets by intercepting their signals rather than using active radar—is essential for stealthy surveillance. Utilizing a swarm of UAVs for this task allows for better spatial diversity and improved localization accuracy through triangulation. However, the transition from theoretical models to real-time 3D implementation is hindered by several practical hurdles.
The primary challenges addressed in this work include:
- Asynchronicity: Sensors on different UAVs often report measurements at irregular intervals, making it difficult to fuse data into a coherent picture.
- Physical and Security Constraints: UAVs must operate within physical limits of speed and turning angles while avoiding collisions with each other, terrain obstacles, or known threat zones.
- Computational Complexity: Optimizing 3D trajectories under these constraints is a non-convex problem. Traditional methods like Genetic Algorithms (GA) are too slow for the split-second decision-making required during flight.
Proposed Methodology
The researchers developed a closed-loop framework designed to optimize UAV trajectories dynamically. The system cycles through three core stages to ensure high-fidelity tracking and safety.
1. Asynchronous Tracking
To handle irregular sensor reports, the framework employs a two-step tracking method. First, a Maximum Likelihood (ML) estimator aggregates asynchronous AOA measurements into "composite measurements" for a given fusion interval. These composite measurements are then processed by a Kalman Filter (KF) to estimate and predict the states of multiple targets.
2. Performance Metric: PC-CRLB
The optimization objective is to minimize the Predicted Conditional Cramér–Rao Lower Bound (PC-CRLB). This metric represents the theoretical floor for tracking error. By moving UAVs into geometries that minimize this bound, the system maximizes sensing accuracy through optimal triangulation angles.
3. Comprehensive Solution Method (CSM)
The CSM is the paper's core algorithmic innovation, designed for speed and robustness:
- Unconstrained Scenarios: When no security constraints are active, the Nonmonotone Spectral Projected Gradient (NSPG) method is used to find optimal paths rapidly.
- Constrained Scenarios: When collision or terrain risks are present, the Alternating Direction Penalty Method (ADPM) is utilized. ADPM decouples complex multi-UAV constraints into simpler, independent subproblems using auxiliary vectors, allowing for efficient parallel computation.
Experimental Evaluation
The framework was tested in high-fidelity 3D simulations involving 4 UAVs tracking 3 targets within a $10km \times 10km \times 2km$ area. The environment included a digital terrain map with 1,600 obstacle points and designated threat zones.
Key Results
- Tracking Accuracy: The CSM achieved performance nearly identical to the computationally expensive Genetic Algorithm, reducing tracking error by over 50% compared to unoptimized linear trajectories in the final stages of the mission.
- Computational Efficiency: The average runtime for the CSM was approximately 0.05 seconds, whereas the Genetic Algorithm required 52.4 seconds. This 1,000x speedup enables real-time onboard implementation.
- Safety and Geometry: The algorithm successfully balanced tracking accuracy with safety, forcing UAVs to maintain safe distances while maneuvering into optimal triangulation positions.
Conclusion
This research demonstrates that real-time, collaborative 3D trajectory optimization for passive tracking is achievable. By decoupling complex constraints and utilizing efficient gradient-based solvers, the CSM provides a scalable solution for multi-UAV systems. While the current "myopic" approach (looking one step ahead) is effective, future work will explore multi-step trajectory optimization to avoid local optima in highly complex environments and integrate real-time obstacle detection sensors. Additionally, addressing communication constraints in contested electronic environments remains a key area for future development.
Summary updated