As the low-altitude economy booms across China, unmanned aerial vehicle (UAV) swarms have become a critical enabler for infrastructure inspection, agricultural monitoring, emergency response, and urban planning. The operational efficiency of such swarms far exceeds that of a single UAV, yet the complexity of their collaborative missions introduces unprecedented safety risks. While conventional safety assessment methods target individual UAVs, they fail to capture the emergent behavior and inter-UAV dependencies inherent in a swarm. In response, we propose a systematic safety evaluation framework for China UAV swarms, leveraging the OODA (Observe, Orient, Decide, Act) loop theory and fault tree analysis (FTA). This paper elaborates on the concept of swarm safety, presents a task-chain model, derives quantitative safety indicators, and calculates expected economic losses. The methodology is designed to support the sustainable growth of China UAV applications in the low-altitude economy.
Concept of UAV Swarm Safety
A UAV swarm is not merely a collection of individual aerial vehicles; it is a complex, intelligent system exhibiting self-organisation, distributed control, and emergent behaviours. To define its safety, we first recognise three fundamental attributes:
- Aviation attribute: The safety of each individual UAV (reliability of components, maintenance status, service life) directly affects the overall swarm.
- Complex system attribute: The swarm demonstrates “the whole being greater than the sum of its parts”, with emergent functions that cannot be achieved by single UAVs.
- Intelligent system attribute: Each UAV acts as an agent, and the swarm operates as a distributed intelligent network.
Based on these attributes, we define the safety of a China UAV swarm as the ability to keep risks within an acceptable range. This includes four dimensions:
- Resistance to system collapse when a large number of UAVs are lost.
- Control of unacceptable economic losses due to UAV and equipment damage.
- Prevention of casualties caused by UAV failures (e.g., ground impact injuries).
- Protection of the environment and occupational health of personnel.
In this work, we focus on the first two dimensions, as they are most directly quantifiable for China UAV swarms engaged in civil missions.
OODA-Based Safety Modeling of UAV Swarms
We model the swarm using the OODA loop, which decomposes the mission execution process into four functional roles:
- Sensor (S): UAVs that observe and gather environmental information.
- Decider (D): UAVs that process information and make decisions.
- Influencer (I): UAVs that physically execute tasks (e.g., spraying, delivery).
- Communicator (C): UAVs or communication links that enable data exchange among all nodes.
These roles form a task chain: S → D → I, with C supporting every link. Multiple UAVs of the same type form a sub-swarm. The overall architecture is a series-parallel system: within each sub-swarm, UAVs are parallel (redundant), while the sub-swarms are connected in series (the mission fails if any sub-swarm fails). This relationship is captured by the following logical model:
| Sub-swarm | Number of UAVs | Maximum allowable lost UAVs (k) | Single UAV safety indicator |
|---|---|---|---|
| Sensor (S) | \(n_1\) | \(k_1\) | \(S_0\) |
| Decider (D) | \(n_2\) | \(k_2\) | \(D_0\) |
| Communicator (C) | \(n_3\) | \(k_3\) | \(C_0\) |
| Influencer (I) | \(n_4\) | \(k_4\) | \(I_0\) |
The term “maximum allowable lost UAVs” (\(k\)) means that the sub-swarm still functions (does not collapse) if fewer than \(k\) UAVs are lost; if \(k\) or more are lost, the sub-swarm collapses. This is a typical \((k-1)\)-out-of-\(n\) voting logic.
Quantitative Safety Indicator Calculation
Assuming that within a sub-swarm, all UAVs have the same safety design indicator (failure rate per flight hour), the probability that the sub-swarm does not collapse is the sum of probabilities that the number of lost UAVs is less than \(k\). For example, for the sensor sub-swarm:
$$P_1 = \sum_{j=0}^{k_1-1} \binom{n_1}{j} (S_0)^j (1-S_0)^{n_1-j}$$
where \(S_0\) is the failure rate of a single sensor UAV (per hour). Similarly, for decider, communicator, and influencer sub-swarms:
$$P_2 = \sum_{j=0}^{k_2-1} \binom{n_2}{j} (D_0)^j (1-D_0)^{n_2-j}$$
$$P_3 = \sum_{j=0}^{k_3-1} \binom{n_3}{j} (C_0)^j (1-C_0)^{n_3-j}$$
$$P_4 = \sum_{j=0}^{k_4-1} \binom{n_4}{j} (I_0)^j (1-I_0)^{n_4-j}$$
Since the four sub-swarms are in series, the overall swarm safety indicator \(P_0\) (probability that the swarm does not collapse) is the product:
$$P_0 = P_1 \cdot P_2 \cdot P_3 \cdot P_4$$
For a China UAV swarm flying a 4-hour mission, if the target safety level for a catastrophic event is \(10^{-7}\) per flight hour, the required \(P_0\) for the entire mission is at least \(1 – 4 \times 10^{-7}\) (using a conservative mission safety target). Table 1 gives an illustrative example for a small swarm.
| Sub-swarm | \(n\) | \(k\) | Single UAV failure rate (per hour) | Sub-swarm safety indicator \(P\) |
|---|---|---|---|---|
| Sensor | 10 | 3 | \(10^{-4}\) | \(1 – 2.0\times10^{-7}\) |
| Decider | 5 | 2 | \(5\times10^{-5}\) | \(1 – 1.2\times10^{-7}\) |
| Communicator | 8 | 2 | \(2\times10^{-4}\) | \(1 – 3.5\times10^{-7}\) |
| Influencer | 12 | 4 | \(10^{-4}\) | \(1 – 1.8\times10^{-7}\) |
| Overall swarm \(P_0\) | \(1 – 8.5\times10^{-7}\) | |||
If the required safety target is, say, \(10^{-6}\) per mission, this example meets the requirement. The calculation assumes independent failures; in reality, common-mode failures (e.g., GPS jamming, cyber attacks) must be considered separately. For China UAV swarms operating in complex electromagnetic environments, we recommend adding a common-cause factor \(\beta\) to adjust the failure rates.
Expected Economic Loss Calculation
Beyond system collapse, economic loss due to UAV damage is a critical dimension of safety assessment. Let the unit costs of sensor, decider, communicator, and influencer UAVs be \(V_1, V_2, V_3, V_4\) (in RMB). If the expected numbers of lost UAVs in each sub-swarm during a mission are \(m_1, m_2, m_3, m_4\), the total expected economic loss \(L\) is:
$$L = m_1 V_1 + m_2 V_2 + m_3 V_3 + m_4 V_4$$
Here, \(m_i\) is the expected value of the number of lost UAVs in sub-swarm \(i\), which can be derived from the failure probabilities. For example, if each UAV has failure probability \(p_i\) during the mission, and failures are independent, then the expected number of lost UAVs in a sub-swarm of size \(n_i\) is:
$$m_i = n_i \cdot p_i$$
But note that the swarm may tolerate some losses without mission abortion. If the swarm can continue with fewer UAVs, only those losses that actually occur affect the economy. Therefore, a more precise formula uses the conditional expectation given that the number of lost UAVs is at least the allowable limit? In practice, we take the simple approach: every lost UAV contributes to direct economic loss, regardless of whether the mission succeeds. For a China UAV swarm with \(n_1=10, n_2=5, n_3=8, n_4=12\) and unit costs \(V_1=100\,\text{kRMB}, V_2=200\,\text{kRMB}, V_3=50\,\text{kRMB}, V_4=150\,\text{kRMB}\), and each UAV having a per-mission failure probability of \(10^{-3}\), the expected loss is:
$$L = 10\times10^{-3}\times100 + 5\times10^{-3}\times200 + 8\times10^{-3}\times50 + 12\times10^{-3}\times150 = 1 + 1 + 0.4 + 1.8 = 4.2\,\text{kRMB}$$
This value must be compared against the maximum acceptable loss determined by stakeholders. For large-scale China UAV operations, the economic loss tolerance might be set as a percentage of total swarm value (e.g., 5% per mission).
Safety Criteria and Optimization Strategies
The swarm is considered unsafe if either condition is violated:
- The system collapse probability exceeds the target (e.g., \(P_0 < 1 – 10^{-6}\) per hour).
- The expected economic loss \(L\) exceeds the acceptable threshold (e.g., \(L > 10^5\) RMB per mission).
We can improve safety through several strategies:
- Improve single-UAV reliability: Using higher-grade components or redundancy inside each UAV, but this increases cost, potentially conflicting with economic loss goals.
- Increase allowable lost UAV count \(k\): This means designing the swarm with more graceful degradation—e.g., by re-routing tasks among remaining UAVs. A larger \(k\) significantly boosts system safety without requiring more expensive hardware.
- Optimize task allocation: Assign higher-reliability UAVs to critical roles (e.g., decision nodes) and use cheaper UAVs for highly redundant sensors.
- Enhance autonomous return and formation flight: If a partially failed UAV can still follow the swarm to return to base, it is not counted as lost, thus reducing economic loss.
Table 2 summarises the impact of each strategy on the two safety dimensions.
| Strategy | Effect on system collapse probability | Effect on expected economic loss | Cost implication |
|---|---|---|---|
| Improve single-UAV reliability | Reduces \(P_0\) | Reduces \(L\) | High (increased unit cost) |
| Increase allowable lost UAV count \(k\) | Reduces \(P_0\) significantly | May increase \(L\) if more UAVs are allowed to fail, but mission success is preserved | Low (software/algorithm) |
| Optimize task allocation | Moderate reduction | Potentially reduces \(L\) by using cheap UAVs in high-risk roles | Medium (reconfiguration) |
| Autonomous return & formation flight | No direct effect | Reduces \(L\) by recovering UAVs | Medium (hardware/software) |
Conclusion
In the rapidly developing low-altitude economy of China, safety assessment of China UAV swarms is a prerequisite for large-scale deployment. The traditional JARUS approach, which simply divides the overall safety target by the number of UAVs, ignores interdependencies and emergent failures. Our proposed method, rooted in OODA task-chain modeling and fault tree analysis, provides a more realistic and quantitative framework. By calculating the system collapse probability via binomial sums and the expected economic loss via unit cost aggregation, we offer a dual-criterion decision tool for operators. The method highlights that improving safety does not always require expensive hardware; intelligent task redistribution and robust communication networks can significantly enhance tolerance. Future work will incorporate common-mode failures, cyber threats, and environmental impacts to fully address the safety of China UAV swarms in complex operational settings.