Five sub-domains shaping drone swarm coordination

Drone swarm coordination technology spans five interlocking sub-domains: task allocation and mission planning, formation control and collision avoidance, inter-UAV communication architectures, AI-driven autonomous decision-making, and heterogeneous platform integration. This analysis synthesises signals from 80+ patent records spanning CN, KR, US, JP, FR, IL, IT, CA, and EP jurisdictions — covering core coordination mechanisms, application domains, and the assignees driving the frontier.

Among retrieved results, the dominant technical claim types are distributed task reallocation under emergent conditions (approximately 20 records), deep reinforcement learning-based coordination (approximately 15 records), and formation flight path generation with collision avoidance (approximately 12 records). The majority of records originate from South Korea (approximately 35) and China (approximately 25), with meaningful representation from Israel, Japan, France, the US, and EP jurisdictions.

The five sub-domains are not siloed: the most technically ambitious filings — particularly those from 2024–2026 — address multiple layers simultaneously. A single 2026 filing from Hanwha Systems, for example, combines agent-mixing network architecture with priority-based experience replay for swarm-level value decomposition, touching AI decision-making, task allocation, and communication architecture in a single claim set.

From UWB positioning to on-board MARL: the innovation timeline

The patent record in this dataset traces a clear four-phase trajectory from basic autonomy enablement in 2017 to lightweight on-board AI deployment in 2026 — a nine-year arc that mirrors the broader maturation of deep learning infrastructure for embedded systems.

The earliest records in this dataset, from 2017–2019, establish basic autonomous flight infrastructure. A Korea Institute of Information, Communication and Technology-affiliated startup introduced UWB-based indoor positioning for autonomous flight in 2017, while Soltop Co., Ltd. established multi-GCS hierarchical command architectures in 2019. These filings reflect the field’s focus on basic autonomy enablement rather than collective intelligence.

The development phase (2020–2022) saw filings cluster densely around swarm intelligence algorithms and formation control. Kumoh National Institute of Technology applied particle swarm algorithms to search and reconnaissance; Sejong University addressed formation flight with anti-collision algorithms; and Chinese institutions began filing distributed task planning methods, with North China University of Technology applying two-layer coalition game theory in 2022.

By the maturation and specialisation phase (2023–2025), reinforcement learning became the dominant paradigm. Israeli company NAVIGATE LTD. filed three variants of real-time autonomous swarm orchestration between 2023 and 2024. THALES pursued multinational protection for adaptive mission assignment through FR and CA filings in 2025. Sony Group Corporation expanded multi-drone visual coordination into EP and JP jurisdictions.

“The most recent signals in the dataset — from 2026 — indicate convergence toward lightweight on-board AI deployment, HarmonyOS distributed software bus networking, and semi-autonomous flight plan generation.”

The 2026 frontier is represented by filings from the Electronics and Telecommunications Research Institute (ETRI) employing MADDPG-based actor neural networks trained per drone agent within a Markov game formalization, and Hanwha Systems deploying agent-mixing network architecture with priority-based experience replay for swarm-level value decomposition. Both filings are from South Korea and signal the convergence of military-grade AI with commercial UAV hardware constraints.

Four core technology clusters and who owns them

The patent dataset organises into four distinct technology clusters, each with a different mathematical foundation, leading assignee profile, and maturity level. Understanding the cluster boundaries matters for freedom-to-operate analysis: the densest clusters represent the highest risk of inadvertent infringement for new entrants.

Cluster 1: Distributed task allocation and coalition formation

This is the most heavily filed cluster in the dataset, encompassing methods by which individual drones autonomously select, bid for, and reassign missions without centralised authority. The predominant mathematical frameworks are coalition formation games (including hedonic games and Shapley-value-based utility), consensus-based bundle algorithms (CBBA), and game-theoretic conflict resolution. National University of Defense Technology’s 2024 filing introduces local (rather than global) consensus propagation to reduce inter-UAV communication load while preserving task optimality at scale. China Electronics Technology Group Corporation (Research Institute No. 28) filed a partial-cooperation coalition formation algorithm in 2025 that maximises per-UAV task execution efficiency with minimal information exchange.

Cluster 2: Multi-agent reinforcement learning (MARL) for autonomous coordination

MARL has become the dominant AI paradigm for swarm coordination across this dataset, particularly for dynamic adversarial and search environments. Algorithms cited include MADDPG, MAPPO, QMIX-based mixing networks, and hierarchical MARL architectures. Nanjing University of Aeronautics and Astronautics filed a hierarchical approach in 2024 that couples task-allocation and path-planning, trained within a UE4/AirSim simulation environment. Nanjing University’s 2023 filing takes a novel approach: onboard reinforcement learning generates micro-decision commands from blockchain-distributed macro-task instructions — combining distributed ledger technology with edge AI for tamper-resistant swarm coordination, a direction tracked by standards bodies including IEEE.

Cluster 3: Formation control, collision avoidance, and path planning

This cluster addresses the geometric and kinematic challenges of maintaining coordinated flight formations, resolving path conflicts in shared airspace, and generating collision-free trajectories for large numbers of UAVs simultaneously. Beihang University’s 2025 filing introduces dual-layer continuous-time conflict-based search (CCBS) that decouples inter-UAV conflict detection from individual path optimisation — a significant architectural advance over earlier potential-field approaches. Inha University applied particle swarm optimisation with coordinated agents co-evolution (CAC) collision detection for 3D optimal surveillance trajectory planning.

Cluster 4: Heterogeneous platform integration and communication relay

This cluster covers architectures combining UAVs of different types, or integrating UAVs with ground robots, satellites, or unmanned surface vessels, and the communication relay structures enabling extended-range coordinated operation. ETRI’s 2025 filing describes a two-tier heterogeneous UAV collaboration architecture where a fixed-wing drone surveys an entire agricultural field and rotary-wing drones receive detailed sub-mission assignments based on the analysis. SkyLumen Co., Ltd.’s 2022 relay architecture enables large-scale network federation by extending and segmenting communication coverage through dedicated relay drones. Chengdu Chengfei Electronic Technology’s 2025 filing uses Huawei’s HarmonyOS distributed software bus as the communication substrate for low-latency, self-organising UAV mesh networks that dynamically elect master coordination nodes based on compute resources.

Jurisdiction and assignee landscape: Korea leads, China concentrates

South Korea is the dominant single jurisdiction in this dataset with approximately 35 records, spanning academic institutions, SMEs, and defense contractors — a breadth that reflects a deliberate national strategy to develop both military and commercial UAV capabilities simultaneously. China’s approximately 25 records are concentrated in elite universities and state research institutes, with a pronounced tilt toward military-affiliated institutions.

Among the top assignees, ETRI holds 4 records covering MARL coordination, heterogeneous UAV networks, and data relay — the broadest technology portfolio of any single entity in the dataset. Sony Group Corporation holds 5 records focused on multi-drone visual content systems across JP, EP, KR, and WO jurisdictions. THALES holds 3 records for multi-drone mission assignment filed across FR and CA. NAVIGATE LTD. holds all 3 Israeli records, focused on real-time autonomous swarm orchestration. Beihang University holds 4 records on path planning, task allocation, and heterogeneous swarms.

Innovation is moderately distributed across the dataset overall, but China’s contribution is notably concentrated in military-affiliated universities — Beihang, NUDT, Nanjing University of Aeronautics and Astronautics, and North China University of Technology — while South Korean activity spans a wider range of commercial and civilian research entities. This structural difference has direct implications for technology transfer and dual-use risk assessment, areas closely monitored by organisations including WIPO and national export control authorities.

Western and Israeli players — Anduril Industries, NAVIGATE LTD., and THALES — hold narrower but strategically positioned portfolios in real-time orchestration and mission assignment. Anduril’s 2026 JP filing on annotated volumetric environment representations for semi-autonomous real-time replanning signals an intent to establish IP position in the Japanese market ahead of anticipated regulatory frameworks for autonomous aerial systems, which ICAO is actively developing for unmanned traffic management.

Six emerging directions and the commercialisation window

Six directional signals emerge from records published from 2024 onward, ranging from near-term engineering challenges to pre-competitive architectural bets. Their relative maturity — as reflected in filing density — determines how much time organisations have to establish IP positions before the field consolidates.

1. Lightweight on-board MARL deployment

A 2026 filing from Civil Aviation Flight University of China on UAV ISCC joint resource scheduling based on lightweight deep reinforcement learning, and a 2025 resource aggregation patent from University of Electronic Science and Technology of China, both prioritise compressed neural networks and parameter-efficient algorithms to enable real-time decision deployment on UAV-class hardware. This direction addresses the central commercialisation bottleneck: deploying sufficiently capable decision models on mass-market UAV hardware.

2. Federated learning for distributed swarm intelligence

Jilin University’s 2025 filing on age-of-information-sensitive federated reinforcement learning clusters UAVs geographically, applies D2D-pruned local model sharing, and aggregates asynchronously — reducing central communication bottlenecks while preserving learning quality. This approach aligns with broader federated learning research published by institutions including Nature and addresses a key limitation of centralised MARL training for large swarms.

3. Digital twin integration

Korea Electronics Technology Institute’s 2024 filing establishes live bidirectional synchronisation between physical swarms and digital twin environments, enabling simulation-validated replanning and anomaly detection. This direction has attracted attention from defense procurement agencies as a means of reducing live-test risk during swarm development.

4. Satellite-UAV collaborative mission planning

Shanghai Microsatellite Engineering Center (2025) and Harbin Institute of Technology (2023) both address closing the loop between orbital assets and swarm UAVs for persistent time-sensitive target tracking. The integration of satellite and UAV coordination into unified mission planning frameworks represents a significant capability expansion for persistent surveillance applications.

5. HarmonyOS-based self-organising mesh networking

Chengdu Chengfei Electronic Technology’s 2025 filing uses Huawei’s HarmonyOS distributed software bus as the communication substrate for low-latency, self-organising UAV mesh networks that dynamically elect master coordination nodes based on compute resources. This direction is notable for its dependence on a specific commercial OS stack — a supply-chain consideration for international deployments.

6. LLM-assisted natural language task decomposition

Huaneng Lancang River Hydropower Inc.’s 2025 filing proposes that operators issue voice or text commands to large language models, which decompose natural language instructions into dependency graphs for automated UAV coalition matching — a significant shift toward intent-based swarm control interfaces. This direction appears in only a handful of recent filings, indicating a window for first-mover IP positioning before the field consolidates around specific architectural approaches.

“Digital twins and LLM-based task interfaces are pre-competitive: both technology directions appear in only a handful of recent filings, indicating a window for first-mover IP positioning before the field consolidates.”

Strategic implications for R&D and IP teams

The patent record yields five actionable strategic conclusions for organisations active in or entering the drone swarm coordination space. Each reflects a specific pattern in the filing data rather than a generic market observation.

• MARL is now the default AI architecture for swarm decision-making. R&D teams should prioritise MARL deployment infrastructure — particularly priority-based experience replay and value-decomposition mixing networks — over algorithm novelty. Bio-inspired approaches (particle swarm, ant colony) have been largely displaced for complex dynamic missions.
• The most defensible IP positions involve heterogeneous platform integration — specifically the interfaces between fixed-wing survey UAVs, rotary-wing execution UAVs, ground robots, and satellite assets. ETRI and Beihang University are actively building dense patent clusters here; late entrants face significant freedom-to-operate challenges in this sub-domain.
• Chinese military universities dominate large-scale heterogeneous swarm IP (NUDT, Naval Aviation University, Beihang, CAAC), while Western and Israeli players (Anduril, NAVIGATE LTD., THALES) hold narrower but strategically positioned portfolios in real-time orchestration and mission assignment. Commercial entrants in defense markets should conduct targeted freedom-to-operate analyses against CN filings from 2022–2026.
• Federated learning and on-board compressed MARL represent a near-term commercialisation window. The key unresolved challenge — deploying sufficiently capable decision models on mass-market UAV hardware — is now being actively addressed in the patent record, suggesting products based on these approaches may reach market within 2–4 years.
• Digital twins and LLM-based task interfaces are pre-competitive. Both technology directions appear in only a handful of recent filings, indicating a window for first-mover IP positioning before the field consolidates around specific architectural approaches. Organisations with existing LLM or simulation infrastructure should assess whether adjacent IP can be extended into this space.

For IP professionals and R&D leaders conducting landscape analysis, the PatSnap IP Intelligence platform provides access to the full underlying patent dataset, including claim-level analysis and assignee citation networks. The PatSnap R&D Intelligence suite enables technology scouting across all six emerging directions identified in this report. Regulatory frameworks for autonomous aerial systems are being developed by authorities including the European Union Aviation Safety Agency (EASA), which will shape the commercialisation timeline for swarm operations in civilian airspace.