Inter-UAV Handoff Protocol

Updated 6 July 2026

Inter-UAV handoff protocol is a set of mechanisms transferring target tracking, identity, or communication roles between drones to maintain operational continuity.
It employs strategies such as vision-based tracking, topology-aware spatiotemporal matching, and cooperative serving-set replacement to overcome dynamic network challenges.
Evaluations demonstrate high target coverage (up to 82.98%) and improved link continuity, highlighting the importance of predictive preparation and rule-based candidate selection.

Inter-UAV handoff protocol denotes the set of mechanisms by which responsibility for a target, identity, route, session, or cooperative serving role is transferred from one UAV to another while continuity is preserved. In current research, the term does not refer to a single standardized packet-level specification. Instead, it spans several technical lineages: vision-based transfer of tracking responsibility between drones, topology-aware identity handover across overlapping UAV fields of view, gateway and mesh architectures that enable path migration, cooperative multi-UAV serving-set replacement, and predictive mobility-management mechanisms that are transferable from UAV-to-infrastructure settings (Kim et al., 17 Jul 2025, Ye et al., 15 May 2026).

1. Scope and problem classes

The literature uses “handoff” to describe several different continuity problems. In some systems, the object being handed off is a visually tracked target; in others it is a global identity, a relay or gateway path, or a cooperative serving set. A concise classification is therefore necessary before protocol details can be compared.

Paradigm	Handoff object	Representative work
Vision-based tracking transfer	Target and tracking responsibility	(Kim et al., 17 Jul 2025)
Spatiotemporal multi-UAV sensing handover	Global vehicle identity across adjacent UAV FOVs	(Ye et al., 15 May 2026)
Gateway/mesh mobility framework	Gateway attachment or relay path	(Roy et al., 2024)
Cooperative multi-UAV communications	Three-UAV serving set	(Li et al., 2024)
Proactive relay/path selection	Direct or relay-assisted path	(Liang et al., 16 Dec 2025)

Directly relevant work is still relatively sparse. “Continuous Marine Tracking via Autonomous UAV Handoff” presents a cooperative two-UAV marine tracking system whose main protocol contribution is a vision-based inter-UAV handoff intended to preserve continuous shark tracking when one drone must leave due to battery limits (Kim et al., 17 Jul 2025). “A Topology-Aware Spatiotemporal Handover Framework for Continuous Multi-UAV Tracking” formulates inter-UAV handoff as identity persistence across a chain of UAVs monitoring a road corridor, using overlap regions, directional constraints, and metadata buffers rather than appearance-based Re-ID (Ye et al., 15 May 2026). “UNet” provides a generic and reliable multi-UAV communication and networking architecture that enables inter-UAV and UAV-to-gateway handoff, but does not define a complete standalone protocol specification (Roy et al., 2024). “Air-to-Ground Communications Beyond 5G: CoMP Handoff Management in UAV Network” studies handoff in a multi-UAV air-to-ground network, where the serving entity is a cooperative set of three UAVs rather than a single UAV (Li et al., 2024).

Several adjacent papers are important as design sources rather than literal inter-UAV protocols. CASH addresses UAV-to-GBS handover in aerial corridors, not peer-to-peer UAV transfer (Saboor et al., 5 Aug 2025). The graph-of-convex-sets URLLC planner and the disconnectivity-aware cargo-UAV planner both treat handoff as inter-BS association change under motion constraints, not UAV-to-UAV reassignment (Ping et al., 17 Nov 2025, Cherif et al., 2021). The SDR LTE testbed demonstrates airborne S1 handover between terrestrial eNBs for a UAV UE, which is valuable methodologically but not an aerial peer-handoff design (Powell et al., 2022). This suggests that the field remains fragmented between direct inter-UAV implementations and transferable mobility-management abstractions.

2. Vision-based transfer of tracking responsibility

A concrete inter-UAV handoff protocol appears in the marine-tracking system of (Kim et al., 17 Jul 2025). The platform uses two quadrotor UAVs, denoted D1 and D2, each with a 425 mm frame, a Pixhawk flight controller, IMU and GPS, an NVIDIA Jetson Nano (8 GB), a 45°-tilted RGB-D camera, Wi-Fi, and ROS Noetic + MAVROS connected to the Pixhawk at 115200 baud. Each UAV runs image capture, deep visual tracking, target state estimation, closed-loop navigation updates, and inter-UAV target-state sharing onboard. High-level vision and handoff logic run on the Jetson Nano, low-level stabilization and fast control run on Pixhawk/Ardupilot, and the design does not depend on cloud processing.

The role split is asymmetric. During normal operation, one UAV is the active tracker and the other is the incoming or replacement UAV. During transfer, the incoming UAV navigates to the vicinity of the active UAV, localizes it, synchronizes with it, receives tracking information over Wi-Fi, performs visual reacquisition, and then takes over. The localization stage uses an ArUco marker mounted on the tracking UAV and GPS to estimate the peer pose

$P = (X, Y, \alpha),$

where $X$ and $Y$ are relative planar coordinates and $\alpha$ is orientation. The incoming UAV aligns parallel to the active UAV while maintaining offsets $X_t$ and $Y_t$ , so that its 45° camera can view both the active drone and the shark region. The paper characterizes this design most accurately as hybrid: distributed in execution, peer-to-peer in transfer, and role-based or leader-follower during handoff.

The handoff workflow is reconstructed from architecture and matching details. The current UAV tracks the shark with OSTrack. The motivating trigger is battery limitation of the current tracker; no numeric battery threshold, timeout, or predictive trigger is given. The receiving UAV then gets real-time location data, navigates to the active UAV, detects the ArUco marker, and initiates what the paper calls a handshake protocol to synchronize the drones. After synchronization, the active UAV wirelessly transmits live tracking data via Wi-Fi. The reacquisition stage uses “high-confidence visual features of the tracked shark,” a two-stage extraction strategy, and ORB feature matching. Specifically, the system first extracts a broad ROI with fixed padding of 300 pixels, then attempts localization with tighter templates. The quantitative table reports template paddings of 30, 50, and 70 pixels. ORB is used for keypoint detection, descriptor extraction, and cross-view matching between the outgoing drone’s template and the incoming drone’s image. The reacquired matched box becomes the incoming UAV’s initial target state, after which the former tracker returns to base.

The protocol is notable for what it omits. It does not publish a formal state machine, exact packet schema, clock synchronization method, acknowledgment logic, numerical confidence threshold, retransmission policy, or failover branch if the incoming UAV fails to reacquire the shark. Its online coordination loop runs at 4–5 Hz, while low-level flight control runs at 100+ Hz. The strongest handoff results are target-centric rather than message-centric: the abstract reports 82.9% target coverage during target transfer attempts via high-confidence feature matching, and the best template setting reaches 82.98% target coverage with a 236-frame longest success streak, corresponding to more than 7 seconds. This establishes feasibility of onboard cross-UAV transfer by appearance matching and relative geometry, but it remains a proof-of-concept onboard visual handoff framework rather than a complete protocol specification.

3. Topology-aware spatiotemporal handoff in multi-UAV sensing networks

A different but direct formulation appears in the multi-UAV traffic-tracking framework of (Ye et al., 15 May 2026), where handoff is the transfer of a vehicle’s global identity from an upstream UAV’s local tracklet to a downstream UAV’s local tracklet. The network is modeled as a directed graph

$\mathcal{G} = (\mathcal{C}, \mathcal{E}),$

with camera/UAV set

$\mathcal{C} = \{C_1, C_2, \dots, C_N\},$

and edges $e_{i,j}$ when adjacent cameras have overlapping ground coverage. Each UAV has a ground-plane projection $\Omega_i$ , and the overlap region is

$X$ 0

A local tracklet state is

$X$ 1

and the handoff problem is written as association of an upstream exiting tracklet to a downstream entering tracklet under spatial, temporal, and directional consistency constraints.

The implementation is explicitly lightweight. Local perception uses YOLO11 and ByteTrack on synchronized 4K streams; the handoff mechanism uses metadata rather than visual embeddings. The architecture has an asynchronous perception layer and a synchronous global consumer. A global synchronization barrier buffers local outputs until a common timestamp is reached, because downstream appearance must not be processed before upstream exit. The effective handoff metadata includes items such as

$X$ 2

and the paper also emphasizes state such as $X$ 3, $X$ 4, and directional zone $X$ 5. Overlap polygons are manually calibrated as road-constrained polygons, and the road is partitioned into directional zones by a road-aligned split

$X$ 6

The handoff logic is a deterministic queue or buffer mechanism. For each directed adjacent pair and directional zone, the system maintains a metadata buffer

$X$ 7

When an upstream track enters the overlap polygon, its metadata is pushed into the corresponding outgoing buffer. When a downstream track is initialized without a global identity, the downstream node queries the appropriate upstream buffer, restricts candidates by temporal window

$X$ 8

and then selects the candidate minimizing lateral displacement: $X$ 9 If the best match is below $Y$ 0, the downstream track inherits the upstream $Y$ 1, and the matched buffer entry is removed. Stale entries older than $Y$ 2 are deleted. The method is therefore FIFO-guided but not strict FIFO; the paper explicitly shows that pure FIFO is insufficient under overtaking and parallel motion, so lateral metadata is required.

This protocol directly challenges the assumption that cross-UAV identity handoff must be appearance-based. In the reported server-side comparison, ByteTrack without topology-aware inter-camera logic achieves HOSR 12.4% and IDF1 34.2%, DeepSORT reaches HOSR 68.3% and IDF1 55.1%, FastReID reaches HOSR 74.1% and IDF1 62.8%, while the proposed method reaches HOSR 99.8%, IDF1 96.5%, and 62.1 FPS. Under traffic regimes, Set A reports HOSR 99.8% with 14.2 ms latency, Set B reports HOSR 92.4% with 18.5 ms latency, and Set C reports HOSR 98.6% with 15.1 ms latency. The ablation in Set C shows that disabling lateral-aware matching reduces HOSR to 74.5%. The protocol therefore demonstrates that, in nadir-view structured traffic scenes, topology-aware spatiotemporal handoff can dominate appearance-based Re-ID. Its limitations are equally explicit: no exact numerical values for $Y$ 3, $Y$ 4, or $Y$ 5, no exact formula for $Y$ 6, no message serialization, and no decentralized peer-to-peer implementation details are provided.

4. Gateway, mesh, and cooperative serving-set handoff

At the networking layer, several papers provide architectures that enable handoff even when they do not publish a complete inter-UAV state machine. “UNet” organizes a multi-UAV system into four layers—UAV Networking Layer, UAV Networking Infrastructure Layer, Data Processing and Service Layer, and Application Layer—and supports UAV-to-UAV, UAV-to-Infrastructure, and Infrastructure-to-Infrastructure communication (Roy et al., 2024). Its most handoff-relevant elements are a hybrid ad hoc plus infrastructure design, a multi-protocol UAV gateway, multimaster ROS-based middleware, and ECMP-assisted gateway-to-gateway transition. The gateway bridges heterogeneous radios and networks using IP forwarding and IP masquerading; UAVcom handles U2U sharing and forwarding; GCScom maintains a virtual connection for data sharing between DPSL and the UAV’s ROS over secure TCP/IP connections. The paper explicitly evaluates gateway handover between $Y$ 7 and $Y$ 8. Reported delays include 0.76 $Y$ 9 0.078 s for TP-Link/UDP $\alpha$ 0, 0.84 $\alpha$ 1 0.1 s for TP-Link/UDP $\alpha$ 2, 1.00 $\alpha$ 3 0.001 s for TP-Link/TCP $\alpha$ 4, and 1.04 $\alpha$ 5 0.082 s for TP-Link/TCP $\alpha$ 6. The architecture therefore supports path migration and service continuity, but it leaves trigger criteria, message formats, dwell timers, and explicit link-quality rules unspecified.

A different communication interpretation appears in the CoMP UAV network of (Li et al., 2024). Here the serving entity is not a single UAV but a cooperative set of three UAVs forming a Delaunay triangle in a Poisson-Delaunay triangulation. The serving set at time $\alpha$ 7 is denoted

$\alpha$ 8

and handoff occurs when the average received power of the current candidate CoMP set is exceeded by that of an adjacent candidate set. Candidate serving groups are triangle vertex triplets, and the selected set is the triangle maximizing the sum of average path-gain terms. To avoid exhaustive search, the paper derives a minimum search radius

$\alpha$ 9

and gives search complexity

$X_t$ 0

This makes handoff a serving-set replacement problem rather than a nearest-node switch. The paper states that each UE has a fixed and unique serving UAV set to avoid real-time dynamic UAV searching and achieve effective load balancing, and simulation shows that the proposed scheme outperforms the conventional Voronoi nearest-serving UAV scheme in coverage probabilities with handoffs, even though the cooperative handoff probability is higher. The implication is that multi-UAV coordination can justify greater handoff complexity when radio continuity improves enough to offset it.

Taken together, these systems broaden the meaning of inter-UAV handoff. In gateway and mesh architectures, the handoff object is a path or mobility anchor. In CoMP systems, it is a serving cluster. In both cases, the decisive issue is less the physical transfer of a target box than the maintenance of continuity across changing UAV adjacency relations.

5. Predictive triggering, mobility coupling, and make-before-break design

A recurring result across the broader UAV mobility literature is that handoff quality depends on anticipatory timing rather than reactive threshold crossing alone. A particularly explicit timing framework is given by the multihomed NEMO design of (Slimane et al., 2012), which is not UAV-specific but is structurally relevant. It decomposes handoff delay as

$X_t$ 1

defines an early preparation trigger lead time

$X_t$ 2

and a later switch-imminent lead time

$X_t$ 3

The design introduces adaptive threshold coefficients satisfying

$X_t$ 4

so that Link_Going_Down starts background L2/L3 preparation and Link_Switch_Imminent performs final path switching before actual link break. It also uses rollback and timeout cleanup. The paper evaluates this as a make-before-break multihomed handoff with standby tunnel establishment. Although it is formulated for NEMO rather than UAV swarms, the protocol pattern—early preparation, late switch, rollback, and limited simultaneous-interface usage—is directly transferable to multi-radio UAV handoff.

Predictive target selection is developed more explicitly in CASH, which studies UAV-to-GBS handover in aerial corridors rather than inter-UAV switching (Saboor et al., 5 Aug 2025). CASH filters candidates by current viability and forward position,

$X_t$ 5

then ranks them by the forward-looking geometric score

$X_t$ 6

Execution remains gated by both relative superiority and absolute serving-link weakness: $X_t$ 7 Because candidate selection is based on future geometric usefulness rather than instantaneous strongest signal alone, CASH reduces handover frequency by up to 78% compared to A3 while maintaining low outage probability. The work is not an inter-UAV protocol, but its forward-only candidate filtering, safety-margin trigger design, and predictive ranking are directly relevant to aerial relay replacement.

Trajectory-coupled formulations make the same point at the motion-planning level. The URLLC-aware planner defines handover times by changes in serving BS index, with total handover count

$X_t$ 8

and minimizes a weighted sum of handovers, path length, and flight time (Ping et al., 17 Nov 2025). The cargo-UAV planner similarly minimizes weighted propulsion energy and handoff count under battery and disconnectivity-rate constraints (Cherif et al., 2021). These are terrestrial-BS association studies, not inter-UAV transfer protocols, but they show that handoff count can be an optimization objective rather than merely an observed KPI. A plausible implication is that inter-UAV handoff decisions should be coupled to planned geometry, expected dwell time, and mission endurance, not only to local instantaneous link measurements.

6. Evaluation practice, recurring limitations, and research directions

The most advanced current systems still expose a consistent boundary between handoff intelligence and handoff protocolization. “Relaying Signal When Monitoring Traffic” proposes a proactive sensing-assisted relay/link selection framework, UAP-Net, in which vehicles fuse LiDAR with multi-UAV RGB features, classify the best candidate path $X_t$ 9, and initiate handoff before current link quality deteriorates (Liang et al., 16 Dec 2025). It reports approximately 10% lower outage probability at a 200 Mbps requirement than a ground-network measurement method and states that the system maintains over 86% of the achievable rate even in the absence of multiple UAVs. Yet the paper is explicit that it does not provide an inter-UAV message sequence, handoff-delay metric, make-before-break mechanism, or control-plane overhead analysis. The result is a strong decision engine, not a full protocol stack.

Evaluation infrastructure is improving faster than protocol closure. The modular 5G connected-UAV simulator of (Su et al., 31 Aug 2025) includes a handoverManager, supports A3, UCB, DDPG, and cross-layer DQN, and integrates TCP, UDP, QUIC, and MP-QUIC. Its DRL state includes candidate SINRs, current-serving indicator, throughput, time since last handover, and cumulative handover count. The SDR LTE platform of (Powell et al., 2022) shows how to stage airborne handover experiments through lab emulation and outdoor testing; it reports that in its setup S1 handover is initiated when the local BS RSRP is 3 lower than that of the neighbor station, and outdoor tests use eNBs approximately 525 m apart with the UAV at 45 m altitude. These works do not solve inter-UAV handoff directly, but they establish a methodological pattern: overlap engineering, candidate measurement before execution, transport-aware logging, and verification from both endpoint and network perspectives.

Across the direct inter-UAV literature, the most conspicuous omissions are still exact message schemas, timer values, retransmission rules, failure branches, and security procedures. The marine tracking system does not specify a packet format, clock synchronization method, numerical confidence threshold, or rollback logic if reacquisition fails (Kim et al., 17 Jul 2025). The topology-aware traffic handoff does not provide exact numerical values for $Y_t$ 0, $Y_t$ 1, $Y_t$ 2, an exact HOSR formula, or decentralized UAV-to-UAV messaging (Ye et al., 15 May 2026). UNet enables path migration but omits formal trigger metrics and association rules (Roy et al., 2024). The CoMP handoff model gives geometry and probability, but not control-plane architecture, synchronization protocol, or cluster-level signaling (Li et al., 2024). This suggests that the field is presently stronger in continuity heuristics, perception logic, and architectural enablers than in standardized end-to-end handoff machinery.

A recurrent misconception is that inter-UAV handoff is already a settled analogue of terrestrial cellular handover. The literature indicates otherwise. What exists today are several partially overlapping protocol families: vision-based target transfer, metadata-based identity continuation, gateway and relay path migration, and cooperative serving-set replacement. Their common design principles are clearer than their common message format. The emerging consensus is that successful inter-UAV handoff depends on predictive preparation, topology restriction, limited candidate sets, onboard execution, and continuity metrics that are specific to the application—target coverage, global identity persistence, outage probability, throughput stability, or cooperative coverage with handoffs—rather than on a single universal trigger or a single universal state machine.