In modern intelligent warfare, UAV drone swarms have become a critical component due to their flexibility, cost-effectiveness, and ability to execute complex missions such as reconnaissance, command, and strike. However, these swarms are frequently subjected to disturbances including electromagnetic interference, anti-drone strikes, and network jamming, which degrade mission performance. To ensure continuous and effective operation, UAV drone swarms must possess a certain degree of resilience—the ability to withstand disruptions and recover functionality. Existing research often neglects the coupling between multi-task payload requirements and communication network topologies. Furthermore, resilience metrics typically lack quantitative characterization of the collapse phase, making them unsuitable for differentiated assessments across mission scenarios. To address these gaps, we propose a comprehensive modeling and resilience assessment framework driven by multi-task payloads. This paper presents a layered communication network model that captures both communication links and task coordination among UAV drones. We then define mission-specific performance metrics for reconnaissance, command, and strike tasks. Finally, we introduce a relay-centric recovery strategy that significantly improves swarm resilience under electromagnetic attacks. Extensive simulations demonstrate that the proposed strategy outperforms traditional recovery methods.
1. Introduction
The rapid development of unmanned aerial vehicle (UAV) technology has enabled UAV drone swarms to perform sophisticated coordinated missions. In practice, these swarms operate in hostile environments where communication links can be disrupted by electromagnetic interference, physical attacks, or environmental obstacles. Resilience, defined as the ability to anticipate, withstand, adapt to, and rapidly recover from disruptions, is essential for ensuring mission success. Traditional network metrics such as connectivity or efficiency fail to capture task-specific performance degradation. Moreover, most existing studies model UAV drone swarms as homogeneous networks, ignoring the heterogeneity introduced by different payloads (e.g., reconnaissance sensors, strike weapons, command relays). This work aims to fill this gap by proposing a multi-task payload-driven modeling approach and a collapse-aware resilience evaluation method.
Our main contributions are: (1) A layered UAV drone swarm communication network model that integrates task payload collaboration. (2) A set of mission-specific performance functions for reconnaissance, strike, and command tasks. (3) A relay-based recovery strategy that prioritizes the restoration of critical relay nodes. (4) A resilience assessment metric that explicitly quantifies the collapse interval, enabling more accurate evaluation of swarm behavior under sustained attacks.
2. Multi-Task Payload-Driven UAV Drone Swarm Model
2.1 Communication Layer
The communication layer is modeled as an undirected unweighted graph \( G(V,E) \), where \( V \) represents UAV drone nodes and \( E \) represents bidirectional communication links. The network evolves from an initial fully connected seed network \( W_0 \) of size \( n_0 \). When a new UAV drone node \( i \) joins, it connects to an existing node \( j \) with probability:
$$
P(i \rightarrow j) = \frac{k_j + \varepsilon}{\sum_{l \in W} (k_l + \varepsilon)} \cdot F(d_{i \rightarrow j})
$$
where \( k_j \) is the degree of node \( j \), \( \varepsilon \) is a bias factor to avoid isolated nodes, and \( F(d_{i \rightarrow j}) \) is the communication distance influence function:
$$
F(d_{i \rightarrow j}) =
\begin{cases}
1, & d_{i \rightarrow j} < r_c \\
\frac{r_c – d_{i \rightarrow j}}{r_c(1-\eta)}, & r_c \leq d_{i \rightarrow j} < r_c/\eta \\
0, & d_{i \rightarrow j} \geq r_c/\eta
\end{cases}
$$
Here \( r_c \) is the communication range, and \( \eta \in (0,1) \) controls distance decay. To prevent over-connectivity, each node’s degree is capped at \( k_{max} \). Each UAV drone locally monitors its neighbors’ degrees and triggers a dynamic reconnection mechanism when the sum of neighbor degrees exceeds a threshold \( \theta \). This adaptive mechanism ensures that the communication topology remains efficient under mission dynamics.
2.2 Task Payload Layer
The task payload layer is a multi-edge network where nodes represent the payloads carried by UAV drones. Each UAV drone in the communication layer maps to exactly one payload node. Links in this layer denote task collaboration relationships: if two UAV drones cooperate on \( q \) tasks, there are \( q \) edges between their corresponding payload nodes. For example, a reconnaissance UAV drone and a strike UAV drone must collaborate to locate and destroy a target, forming multiple edges representing different sub-tasks (e.g., target identification, handover, attack authorization).
2.3 Interlayer Coupling
The coupling between the communication layer and the task payload layer is one-to-one. When two UAV drones in the communication layer are connected (directly or via relays), their corresponding payload nodes can form task edges. Conversely, if a communication link fails, all task edges involving the failed UAV drone are also disabled. This dependency captures the realistic scenario where mission coordination relies entirely on communication connectivity.
3. UAV Drone Swarm Performance Functions
We define three fundamental factors that characterize swarm operational capability: area coverage, focus, and time endurance. These factors are combined to form mission-specific performance metrics.
3.1 Area Factor
Each UAV drone has a circular coverage area with radius \( r_c \). The total effective coverage of a swarm of \( m \) UAV drones, considering communication status \( C_i \in \{0,1\} \), is the union of individual areas:
$$
S(t) = \bigcup_{i=1}^m C_i S_i(t)
$$
The area factor is defined as the ratio of current coverage to initial coverage:
$$
R_S(t) = \frac{S(t)}{S(t_0)}
$$
When all UAV drones fail, \( R_S(t)=0 \); when all are operational, \( R_S(t)=1 \).
3.2 Focus Factor
Focus measures the degree of overlapping coverage among multiple UAV drones. Let \( A(t) \) be the area covered by at least two UAV drones simultaneously. The focus factor is:
$$
R_A(t) = \frac{A(t)}{\sum_{i=1}^m \pi r_c^2}
$$
When coverage is completely disjoint, \( R_A(t)=0 \); when fully overlapped, \( R_A(t)=1 \).
3.3 Time Factor
The time factor quantifies the endurance of a specific payload. Let \( L \) be the active duration of a payload within a total mission time \( T \). Then:
$$
R_T = \frac{L}{T}
$$
\( R_T \in [0,1] \).
3.4 Reconnaissance Performance
Reconnaissance missions aim to collect maximum information from the target area. The performance depends on area coverage and endurance, weighted by network connectivity:
$$
P_C(t) = \frac{R_S^C(t) \cdot R_T^C}{L_S(t)}
$$
where \( R_S^C(t) \) is the area factor of reconnaissance UAV drones, \( R_T^C \) is their time factor, and \( L_S(t) \) is the network connectivity metric (e.g., giant component ratio).
3.5 Strike Performance
Strike missions require precise coordinated attacks. Performance depends on area coverage, focus, and network efficiency:
$$
P_D(t) = \frac{R_S^D(t) + R_A^D(t)}{2} \cdot X(t)
$$
where \( X(t) = \frac{1}{N(N-1)}\sum_{i \neq j} \frac{1}{d_{ij}(t)} \) is the network efficiency, measuring the average inverse shortest path length.
3.6 Command Performance
Command missions emphasize timely message delivery. The performance is:
$$
P_Z(t) = \frac{\sum_{i=1}^{N(t)} \delta_{d_i} \cdot M_i(t)}{\sum_{i=1}^{N} M_i(t_0)} \cdot R_T^Z
$$
where \( \delta_{d_i} \) is a time-decay factor for message value, \( M_i(t) \) is the number of messages received by UAV drone \( i \), and \( R_T^Z \) is the command payload endurance factor.
4. Attack and Recovery Models
4.1 Integrated Attack Model
We propose a unified attack model that covers four types of disruptions: random electromagnetic interference, central node attack, relay node attack, and local node attack. The probability of UAV drone \( i \) being attacked in phase \( l \) is:
$$
w_{i,l}(\psi,\xi) = \frac{ \xi \cdot (1-\xi)^{D_{il}} }{ \sum_{j \in V_l} \xi \cdot (1-\xi)^{D_{jl}} }
$$
where \( \psi \in \{0,1,2,3\} \) selects the topology feature \( D_{il} \):
- \( \psi = 0 \): \( D_{il} = 0 \) (equal probability for all nodes)
- \( \psi = 1 \): \( D_{il} = k_{il} \) (degree-based, targets hubs)
- \( \psi = 2 \): \( D_{il} = b_{il} \) (betweenness-based, targets relays)
- \( \psi = 3 \): \( D_{il} = c_{il} \) (clustering-based, targets dense subgraphs)
Here \( \xi \) is a bias parameter. This model unifies random and targeted attacks into a single mathematical framework.
4.2 Failure Process
When a communication layer UAV drone fails, its corresponding payload node also fails, removing all task edges. Additionally, any UAV drone that becomes isolated (no communication links) is considered failed. This cascading effect propagates through the layered network.
4.3 Relay-Centric Recovery Strategy
Traditional recovery methods often prioritize restoring high-degree nodes. However, we observe that in mission-oriented swarms, relay nodes (with high betweenness centrality) are critical for re-establishing connectivity. The importance of UAV drone \( i \) in phase \( k \) is defined as:
$$
I_{i,k} = \frac{\sigma_{st}(i)}{\sum_{s \neq t \in V_k} \sigma_{st}}
$$
where \( \sigma_{st}(i) \) is the number of shortest paths between \( s \) and \( t \) that pass through \( i \). Our recovery strategy sorts surviving UAV drones by \( I_{i,k} \) in descending order and sequentially establishes new communication links within their range. This prioritizes the restoration of bridging nodes, rapidly rebuilding the network backbone.
5. Resilience Assessment Metric
We adopt the resilience definition from literature:
$$
R = \alpha \delta_d \sigma_d \rho_d + \beta \delta_r \sigma_r \rho_r
$$
where \( \alpha + \beta = 1 \), \( \alpha, \beta \in [0,1] \). The subscript \( d \) denotes the degradation phase and \( r \) the recovery phase. Each phase has three factors:
| Phase | Process factor \( \delta \) | State factor \( \sigma \) | Time factor \( \rho \) |
|---|---|---|---|
| Degradation | \( \delta_d = \frac{\int_{t_d}^{t_r} y(t) dt}{(t_r – t_d) y_0} \) | \( \sigma_d = \frac{y_{min}}{y_0} \) | \( \rho_d = \frac{\Delta}{B}(t_r – t_d) \) |
| Recovery | \( \delta_r = \frac{\int_{t_r}^{t_s} y(t) dt}{(t_s – t_r) y_0} \) | \( \sigma_r = \frac{y_s}{y_0} \) | \( \rho_r = \frac{\Delta}{B}(t_s – t_r) \) |
Here \( t_d \), \( t_r \), \( t_s \) are times of attack, recovery start, and stabilization; \( y_0 \), \( y_{min} \), \( y_s \) are initial, minimum, and steady-state performance values; \( \Delta \) is a time weight (\( 0 < \Delta \leq 1 \)); \( B \) is a reference time. The collapse interval is explicitly included in \( y_{min} \) and the integral, enabling quantitative characterization of severe performance drops.
6. Experimental Results
6.1 Simulation Setup
We simulate a UAV drone swarm of 100 nodes using parameters in Table 2.
| Parameter | Value | Description |
|---|---|---|
| \( \alpha \) | 0.4 | Weight for degradation |
| \( \beta \) | 0.6 | Weight for recovery |
| \( r_c \) | 50 | Communication range |
| \( \varepsilon \) | 0.4 | Bias factor |
| \( \eta \) | 0.6 | Distance decay |
| \( \Delta \) | 0.8 | Time weight |
The network is generated with the proposed probability and distance functions. The degree distribution follows an exponential form \( p(k) = 0.43 e^{-0.36k} \), confirming the scale-free property with moderate heterogeneity.
6.2 Performance Sensitivity
We compare the proposed performance functions with traditional metrics (coverage area and network efficiency) under random electromagnetic interference. As shown in Figure 11 (conceptual), our functions exhibit higher sensitivity to node failures, especially during early attack stages. This allows earlier detection of critical node losses.
6.3 Resilience Under Different Attacks
We evaluate resilience for reconnaissance, strike, and command tasks under four attack types. The proposed relay-centric recovery strategy is compared with the baseline strategy from literature (prioritizing hub recovery) and a random recovery strategy.
| Mission | Random interference | Central attack | Relay attack | Local attack |
|---|---|---|---|---|
| Reconnaissance | +9.2% | +42.9% | +7.9% | +57.0% |
| Strike | +24.3% | +16.9% | +24.7% | +18.9% |
| Command | +14.4% | +21.4% | +12.8% | +17.5% |
Table 3 shows that the relay-centric recovery strategy consistently outperforms the baseline. Notably, in the reconnaissance task under central attack, resilience improves by 42.9% because reconnaissance networks rely heavily on relay nodes for global information fusion. In strike tasks, the improvement is substantial under random interference (24.3%) as the strategy rapidly restores coordination among dispersed strike units.
6.4 Collapse Interval Analysis
We also examine the collapse interval (attack ratio > 50%). Table 4 presents resilience values for attack ratios from 60% to 100% under random interference.
| Mission | Recovery type | 60% | 70% | 80% | 90% | 100% |
|---|---|---|---|---|---|---|
| Reconnaissance | Random | 0.0227 | 0.0222 | 0.0073 | 0.0020 | 0 |
| Reconnaissance | Proposed | 0.0445 | 0.0209 | 0.0100 | 0.0040 | 0 |
| Strike | Random | 0.0685 | 0.0551 | 0.0343 | 0.0060 | 0 |
| Strike | Proposed | 0.0689 | 0.0624 | 0.0412 | 0.0247 | 0 |
| Command | Random | 0.0859 | 0.0477 | 0.0170 | 0.0025 | 0 |
| Command | Proposed | 0.0944 | 0.0450 | 0.0170 | 0.0039 | 0 |
In the collapse interval, resilience values are very low (< 0.1) as the network is heavily fragmented. The proposed strategy still offers marginal improvements, but the gap narrows because few operable UAV drones remain for reconnection. This underscores the importance of early detection and recovery before network collapse.
7. Conclusion
We have presented a comprehensive framework for modeling and assessing the resilience of multi-task payload-driven UAV drone swarms. The layered network model captures the realistic coupling between communication and task coordination. The mission-specific performance functions (reconnaissance, strike, command) provide fine-grained evaluation of operational effectiveness under disturbances. The proposed relay-centric recovery strategy demonstrates significant improvements over existing methods, with average resilience gains of 16.5%, 29.3%, and 21.2% for reconnaissance, strike, and command tasks, respectively. The collapse-aware resilience metric offers a quantitative basis for designing robust swarm architectures. Future work will explore intelligent adaptive recovery algorithms that combine real-time topology data with deep reinforcement learning to further enhance swarm autonomy in contested environments.