Technical Summary: Collaborative Threat-Aware Autonomy (CTAA)

1. Problem Statement

The paper addresses the fundamental reliability challenge of navigating unmanned vehicle teams through contested airspace containing dynamic, kinematically constrained hostile pursuers. While recent advances in single-vehicle threat-aware autonomy (specifically using Weapon Engagement Zones or WEZs) allow a vehicle to reactively avoid capture, a single vehicle remains a "single point of failure." If an individual vehicle has a success probability $p$ (typically 0.65–0.75 in challenging scenarios), the mission fails with probability $1-p$. The authors argue that current state-of-the-art approaches focus on minimizing individual risk or path length but fail to optimize for team-level mission success probability, particularly in scenarios where multiple vehicles must transit dynamic WEZs.

The core problem is formulated as a multi-agent trajectory planning task where a fleet of Autonomous Collaborative Platforms (ACPs) must traverse a corridor containing overlapping WEZs generated by turn-constrained pursuers (modeled as Dubins vehicles). The objective is not merely to minimize the exposure of every individual vehicle, but to maximize the probability that at least one mission-critical vehicle reaches its goal, even if lower-value vehicles are captured.

2. Methodology

The proposed framework, Collaborative Threat-Aware Autonomy (CTAA), utilizes a role-differentiated multi-agent architecture combined with independent reactive guidance.

A. Role Differentiation and Spatial Separation

The team consists of three distinct roles, each assigned a specific weight ($w_i$) reflecting mission value:

• Primary Intercept ($w=2.0$): The high-value vehicle tasked with mission success.
• Escort Support ($w=1.0$): A medium-value vehicle.
• Decoy Alternate ($w=0.5$): A low-value vehicle.

These roles are assigned to spatially separated routes (lateral offsets of $\pm 0.9$ m). This separation ensures that the vehicles are not simultaneously observable by a single threat, breaking the "single point of failure" topology.

B. Reactive Guidance: CSBEZ

Each ACP operates independently without inter-vehicle communication. They utilize a Collision Sphere Boundary for Evader Zero-Set (CSBEZ) reactive controller.

• Mechanism: The CSBEZ replaces the infinite-turn-rate Basic Engagement Zone (BEZ) with a model accounting for the pursuer's minimum turn radius (Dubins arc-plus-straight path).
• Control Law: At each time step, the controller sweeps 72 candidate headings to find the worst-case safety margin ($z_{min}$). It then blends the gradient of this safety margin with the direction toward the goal to generate a steering command.
• Independence: Each vehicle calculates its own path based on its state and the shared pursuer parameters. No coordination logic is required between agents.

C. Team-Level Metrics

The authors introduce a suite of metrics to evaluate team performance:

• Cumulative WEZ Exposure ($J_{WEZ}$): Role-weighted sum of risk exposure.
• Violation Score ($V_{WEZ}$): Exposure above a risk threshold.
• Safety Margin ($m_{team}$): Minimum safety margin across the team.
• Mission Success Probability ($P_{mission}$): Defined analytically as $1 - (1-p)^N$ under the assumption of independent outcomes, where $p$ is the individual success rate and $N$ is the number of vehicles.

3. Key Contributions

The paper presents five specific contributions:

1. Role-Differentiated Framework: A multi-ACP architecture where primary, escort, and decoy vehicles run independent CSBEZ controllers on spatially offset routes.
2. Team Metric Suite: A set of metrics ($J_{WEZ}$, $V_{WEZ}$, $m_{team}$, $P_{mission}$) that capture the trade-off between individual risk and team-level outcomes.
3. Deterministic Validation: Simulation results on a "two-gate gauntlet" scenario showing a +35.8 percentage-point improvement in mission success probability (from 0.72 for a single vehicle to 0.978 for a three-vehicle team).
4. Monte Carlo Validation: 100 trials with perturbed threat parameters (range, position, speed) confirming an empirical team success rate of 1.00, matching the analytical redundancy bound.
5. Reactive Threat Demonstration: Experiments with PurePursuit adversaries (which dynamically target the nearest vehicle) demonstrating that the escort and decoy naturally absorb the threats, leaving the primary uncontested without any explicit communication or coordination.

4. Results

The study evaluated three cases: a direct baseline (no threat awareness), a single CSBEZ-aware vehicle, and the three-vehicle CTAA team.

• Deterministic Simulation:
  • The direct baseline was captured immediately.
  • The single CSBEZ vehicle succeeded but incurred a 41% path-length penalty and faced a peak WEZ risk of 0.68.
  • The CTAA team achieved a 100% mission success rate (2/3 vehicles reached the goal). Crucially, the primary vehicle's peak risk dropped to 0.52 (a reduction from the single-vehicle case) because the escort was captured, effectively "saturating" the threat and clearing the corridor for the primary.
• Monte Carlo Validation:
  • Single-vehicle success rate: $0.66 \pm 0.09$ (consistent with the analytical prediction of 0.72).
  • Multi-ACP team success rate: $1.00 \pm 0.00$ across 100 trials, validating the redundancy model.
• Reactive Threat Experiment:
  • When threats were programmed to target the nearest ACP (PurePursuit), the escort and decoy were captured within 2.5 seconds.
  • The primary vehicle transited the now-uncontested corridor with zero WEZ risk, achieving a success probability of 1.00. This occurred purely through passive geometric deception via route separation.

5. Significance and Claims

The paper claims that CTAA successfully bridges the gap between single-vehicle threat avoidance and team-level mission success. The significance lies in demonstrating that role assignment and spatial route diversity can improve mission reliability through two complementary mechanisms:

1. Probabilistic Redundancy: Simply having $N$ independent paths raises success probability to $1-(1-p)^N$.
2. Threat Saturation: Lower-priority vehicles (escorts/decoys) can draw the attention of capacity-limited threats, effectively "sacrificing" themselves to reduce the risk exposure of the high-value primary vehicle.

The authors emphasize that this approach requires no inter-vehicle communication, relying instead on the emergent behavior of independent reactive controllers operating on separated routes. The results suggest that for contested airspace missions, mission reliability is best achieved not by making every vehicle invulnerable, but by structuring the team to ensure that at least one critical asset survives the engagement.

Note: The paper acknowledges limitations, noting that the current risk model is a logistic surrogate rather than a first-principles derivation, and that future work will address moving targets, joint pre-planning, and dynamic role re-assignment.