Pith. sign in

REVIEW 3 major objections 5 minor 90 references

Towards Effcient Low Altitude Sensing: A Dual Heterogeneous Graph Learning Method for UAV Task Allocation

T0 review · 3 major / 5 minor · reviewed 2026-07-11 · grok-4.5

Pith's one-line read Matching a directed task-dependency graph to a UAV communication graph with graph attention and PPO yields more completed multi-UAV sensing tasks in less time.

desk verdict Solid dual-graph GAT+PPO allocator for dependent multi-UAV sensing/compute: real engineering depth, consistent ablations, narrow N=5 sims—still referee-worthy. read the letter →

arxiv 2607.04255 v1 pith:KLKTDT73 submitted 2026-07-05 eess.SY cs.SY

classification eess.SYcs.SY
keywords unmannedaerialvehiclestaskallocationheterogeneousgraphlearningattentionnetworksproximalpolicyoptimizationdependencymulti-UAVsensingnaturallanguagemissionplanning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Complex low-altitude sensing missions are not bags of independent targets: subtasks have order and data dependencies, and UAVs must fly, sense, talk, and share compute under resource limits. This paper claims that the right move is to build a directed task graph for those dependencies and resource needs, an undirected UAV graph for communication and capabilities, and treat allocation as structural matching between the two. Graph attention networks learn node embeddings from both graphs; a cross-attention policy trained with proximal policy optimization jointly chooses which UAV senses each ready task, which links to keep, and how to split compute. In simulation that joint dual-graph policy finishes more tasks with shorter system completion time than distance, resource, random, and ablated baselines across task counts and dependency densities, and an AirSim case shows a language model can emit the task graph from plain-language instructions so the same policy can run a full sensing-and-detection mission.

What carries the argument

Dual heterogeneous graph structural matching: a directed task graph and undirected UAV graph whose GAT embeddings are cross-attended and fed to PPO heads that jointly output assignment Y, topology A, and compute-split X, aided by SCA resource allocation and a pretrained completion-time predictor that ranks candidate decisions.

What would settle it

Repeat the same task-count and complexity sweeps with a larger fleet, intermittent link failures, and tasks that appear online instead of as a fixed known DAG; if dual-graph PPO no longer beats greedy and mean-split on completed tasks and makespan under matched seeds, the central claim does not hold for the regimes the paper targets.

Watch

Extended reading notes

Core claim

Multi-UAV sensing-and-computing allocation under precedence constraints is best cast as structural matching between a directed acyclic task graph and an undirected UAV communication graph. Encoding both with graph attention message passing and deciding assignment, topology, and compute splitting jointly via cross-attention and proximal policy optimization produces higher successful-task counts and lower makespan than the paper’s benchmarks under varying task numbers and graph complexities.

Load-bearing premise

The claimed gains rest on a small, fully observed simulated fleet—five UAVs, few tasks, a known fixed task graph, and idealized rates and queues—being enough like real low-altitude operations that dual-graph matching still wins when links fail, tasks arrive online, or the swarm grows.

Editorial extensions

If this is right

  • Joint dual-graph matching raises completed tasks and lowers system makespan as task count and dependency density increase, relative to no-split, mean-split, GNN-PPO, MLP-PPO, and greedy baselines.
  • About five message-passing rounds best capture multi-hop task dependencies and link quality before oversmoothing hurts decisions.
  • A pretrained completion-time predictor can rank a small set of candidate decisions and improve realized makespan without replacing the policy.
  • Natural-language mission text can be turned by a language model into a structured task graph that the same dual-graph policy then executes for multi-UAV sensing, offload, and detection.
  • Connectivity-aware topology plus compute splitting keeps the swarm graph usable while spreading queue load.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If task-versus-agent graph matching is the right inductive bias, the same pairing may help multi-robot warehouse or disaster-response allocation where jobs form DAGs and robots share a dynamic radio mesh.
  • Scaling past the paper’s small N and M will likely need hierarchical or sampled message passing so attention cost does not explode with every new UAV–task pair.
  • Online task arrivals (rather than a fixed known DAG) would test whether cross-attention PPO can re-match without full re-encoding from scratch.
  • The language-model-to-task-graph interface suggests operators could specify missions in chat and still obtain optimizable, verifiable dependency structure instead of free-form action scripts.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper formulates multi-UAV sensing and computing task allocation as structural matching between a directed task-dependency DAG and an undirected UAV communication graph. Task and UAV attributes, precedence constraints, rates, queues, CBF-safe motion, and SCA-based resource allocation are modeled in detail. GAT message passing extracts structural embeddings from both graphs; cross-attention with PPO jointly decides assignment Y, topology A, and compute split X, assisted by a pretrained completion-time prediction head that ranks K candidate decisions. Simulations (N=5, M up to 8, varying graph complexity CT) report higher successful-task counts and lower makespan than No-Split, Mean-Split, GNN-PPO, MLP-PPO, Random, and Greedy. An AirSim case uses an LLM to convert natural-language requirements into a task graph for closed-loop execution.

Significance. If the dual-graph matching advantage holds beyond the reported regime, the work offers a useful systems contribution for low-altitude collaborative sensing: it unifies task precedence and UAV topology in one state representation, jointly optimizes assignment/topology/split, and provides a concrete LLM-to-task-graph interface. Strengths include a carefully specified system model (DAG ready sets, FCFS queues, SCA residual reallocation, CBF projection), systematic ablations of message-passing depth and predictor architecture, and an end-to-end AirSim demonstration. The contribution is primarily empirical and architectural rather than theoretical; its lasting value depends on whether the reported gains survive larger fleets, imperfect communication, and online task arrival.

major comments (3)
  1. §IV.A and Table 1: The central superiority claim (Abstract; Figs. 5, 7) is established only for N=5 UAVs, M typically 1–8, a 100 m×100 m area, fully known DAGs, idealized rates/queues, and fixed evaluation seeds. The model already includes ready sets M_ready, connectivity λ2(L)>0, and link rates, yet evaluation never stresses packet loss, online task arrival, N≥10, or statistical variability (no error bars/confidence intervals). Without at least one scaled or imperfect-communication experiment, the dual-graph matching advantage is not shown to transfer beyond the toy fleet that underpins the paper’s strongest claim.
  2. §III.C–E and reward (60): Joint decisions (Y,A,X) plus a pretrained predictor that selects among K=4 candidates are load-bearing for the reported gains, but the manuscript does not isolate the predictor’s contribution (e.g., PPO alone vs. PPO+predictor ranking) or report sensitivity of results to K, ω1–ω3, or η1–η2. Given that reward terms align with the evaluation metrics by design, a controlled ablation of the prediction/reconfiguration module is needed to attribute performance to dual-graph matching rather than to the evaluator or reward shaping.
  3. §IV.B AirSim case: The LLM→task-graph→execution pipeline is presented as evidence of engineering potential, yet Fig. 9 only compares Proposed vs. Random on makespan under varying CT, with no quantitative comparison to the paper’s main baselines, no failure modes of LLM graph construction, and no report of detection/compute fidelity under the generated splits. The application case currently supports feasibility more than the dual-graph claim.
minor comments (5)
  1. Title and abstract: “Effcient” is misspelled; abstract/intro also alternate “UAV”/“UA V” spacing inconsistently.
  2. Fig. 3 caption and §III.B: Cross-attention / multi-head interaction is described at a high level; a short equation or pseudocode for the UAV–task attention heads would improve reproducibility.
  3. Table 1: N0 = 0.018 W is unusually large for noise power spectral density units; clarify whether this is N0·B or a simulation-scaled constant.
  4. Benchmark naming is inconsistent across text and figures (Mean Split / Uniform Split; No Split / No-Split). Align labels with §IV.A.1.b.
  5. Several related-work citations are arXiv preprints or forthcoming; ensure final versions and page ranges are updated for production.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: empirical dual-graph GAT+PPO allocation paper; reward matches evaluation by design (standard RL), not a tautological derivation.

full rationale

This is a systems/RL paper that formulates multi-UAV sensing-computing allocation as structural matching between a directed task DAG and an undirected UAV communication graph, encodes both with GAT-style message passing, and optimizes assignment, topology, and compute split via PPO with a cross-attention policy. The claimed superiority is purely empirical (more completed tasks, shorter makespan vs No-Split, Mean-Split, GNN-PPO, MLP-PPO, Greedy, etc. under varying M and CT). The terminal reward r=ω1 Nsucc−ω2 Ttotal−ω3 Pconn deliberately aligns with the reported metrics—normal for PPO and not a self-definitional reduction of a first-principles prediction. The completion-time prediction head is pretrained on simulator labels and used only as a candidate selector (K=4), not presented as an independent physical forecast forced by a fitted constant. No uniqueness theorem, load-bearing self-citation chain, or ansatz smuggled via prior author work underpins the central claim. Self-citations in the bibliography are background (localization, co-design, LLM-UAV surveys) and are not required for the dual-graph matching result. Against external baselines and ablations the derivation is self-contained; score 0 is the honest finding.

Assumptions & free parameters 6 free parameters · 6 assumptions · 2 invented entities

Claims rest on standard wireless/control models, a DAG task model, RL training in a small custom simulator, and many hand-chosen weights and architectural knobs. No new physical entity is postulated; the dual heterogeneous graph is a modeling construct. Free parameters dominate the ledger because performance is learned and tuned rather than derived.

free parameters (6)
  • Reward weights ω1, ω2, ω3 (and objective weights η1, η2)
    Trade off completed tasks, makespan, and connectivity; values not fixed by theory and directly shape the reported optimum.
  • Message-passing depth R (selected around 5)
    Architectural hyperparameter chosen from ablation; performance peaks then degrades at 6 rounds.
  • Candidate decision count K (=4)
    Predictor re-ranking budget fixed by authors; affects selected policy quality.
  • PPO clipping ε_ppo, entropy β, loss coefficients c_v, c_p
    Standard RL knobs that control update stability and the influence of the prediction head.
  • Simulation resource ranges (f_i, B_max, P_max, L_m, c_m, τ_max, etc.)
    Scenario generators define the distribution on which superiority is measured (Table 1).
  • SCA residual reallocation weights and step ρ
    Heuristic resource refinement after topology/assignment; not uniquely determined.
assumptions (6)
  • domain assumption Task dependency structure is a known directed acyclic graph; a task is ready only after all predecessors complete (Eqs. 3–4, 18).
    Load-bearing modeling choice; cyclic or partially observed dependencies would break the ready-set logic.
  • domain assumption Link rates follow Shannon formula with equivalent LoS-weighted channel gain; power/bandwidth budgets and R_min apply (Eqs. 11–13).
    Standard wireless abstraction used for transmission times and topology feasibility.
  • domain assumption UAV motion is a double integrator with PD nominal control projected under CBF safety constraints (Eqs. 14–17).
    Separates flight feasibility from allocation learning; flight times use average speed to waypoints.
  • domain assumption Each active task is assigned to exactly one sensing UAV; compute may be split over local/neighbor nodes with FCFS queues (Eqs. 19–22, 20).
    Defines completion time and success indicator χ_m used in reward and metrics.
  • domain assumption Retained communication topology must remain algebraically connected (λ2(L)>0) (constraints 23c, reward P_conn).
    Hard structural constraint and penalty term in learning.
  • ad hoc to paper GAT/attention message passing plus PPO can learn near-optimal joint assignment/topology/split policies from episodic simulation.
    Methodological premise of the solution framework; not proved, only empirically supported in the authors’ simulator.
invented entities (2)
  • Dual heterogeneous graph G = (U, M, E_UU, E_TT) as the unified state for structural matching
    purpose: Encode task dependencies and UAV communication/resources in one object so allocation is graph matching rather than isolated bipartite assignment.
    Modeling construct, not a new physical object; independent evidence is only empirical performance in simulation.
  • Completion-time prediction head used as pretrained evaluator over K candidate decisions
    purpose: Re-rank policy samples by predicted makespan without generating decisions directly.
    Architectural component trained on the same simulator labels; no external validation dataset.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Effcient Low Altitude Sensing: A Dual Heterogeneous Graph Learning Method for UAV Task Allocation." pith.science (2026). https://pith.science/paper/KLKTDT73

@misc{pith2026260704255,
  author       = {Pith},
  title        = {Pith review of: Towards Effcient Low Altitude Sensing: A Dual Heterogeneous Graph Learning Method for UAV Task Allocation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KLKTDT73}},
  note         = {Machine review of arXiv:2607.04255}
}
read the original abstract

With the development of low altitude intelligent systems, multiple unmanned aerial vehicles (UAVs) can collaboratively execute more complex tasks. Conventional task allocation methods usually regard tasks and UAVs as isolated entities, making it difficult to capture task dependencies and UAV communication relationships. To address this issue, this paper proposes a dual heterogeneous graph learning based UAV task allocation method. A directed task graph is constructed to represent task dependencies and encode task resource requirements, while an undirected UAV communication graph is built to model communication relationships and encode UAV resource states. The task allocation problem is formulated as a structural matching problem between the task graph and the UAV communication graph. A graph attention network based feature extraction method is introduced to learn structural representations from both graphs through message passing. A cross attention mechanism is further integrated with proximal policy optimization to optimize the matching between task nodes and UAV nodes for task allocation. Simulation results demonstrate that the proposed method achieves a higher task completion rate and shorter task completion time than benchmark methods under different evaluation settings. Furthermore, a UAV sensing and computing application is developed on the AirSim simulation platform. A large language model is employed to convert natural language task requirements into a structured task graph for autonomous UAV task execution, demonstrating the potential of the proposed framework for natural language driven UAV mission planning and execution.

Figures

Figures reproduced from arXiv: 2607.04255 by the authors.

Figure 1
Figure 1. Low altitude sensing and computing scenario with multiple UAVs, task dependency relations and sensing areas. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Task dependency graph examples with increasing graph complexity. Yellow nodes denote tasks without predecessor constraints and white nodes denote [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Overall architecture of the proposed dual heterogeneous graph learning framework. UAV and task graphs are encoded by graph attention and directed [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Reward convergence under different architectural choices. [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: Task completion performance under changing task number and task graph complexity. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: Completion time prediction accuracy under varying task graph [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: Impact of message passing rounds on task execution and prediction performance. [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: Application case of task graph guided multi UAV collaborative execution from natural language instructions. [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: Makespan versus graph complexity in AirSim simulation. [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

90 extracted references · 5 linked inside Pith

  1. [1]

    Communication and Control Co-Design for UA V Trajectory Tracking: An Event-Triggered Strategy,

    Q. Liang, Y . Ping, T. Liang, and T. Zhang, “Communication and Control Co-Design for UA V Trajectory Tracking: An Event-Triggered Strategy,” in2025 IEEE 102nd Vehicular Technology Conference (VTC2025-Fall), 2025, pp. 1–5

  2. [2]

    Adaptive Communication and Control Co-Design for Remote UA V Systems,

    H. Ding, T. Liang, Y . Ping, and T. Zhang, “Adaptive Communication and Control Co-Design for Remote UA V Systems,” in2025 IEEE 101st Vehicular Technology Conference (VTC2025-Spring), 2025, pp. 1–6

  3. [3]

    UA V-Aided Positioning Systems for Ground Devices: Fundamental Limits and Algorithms,

    T. Liang, T. Zhang, J. Yang, D. Feng, and Q. Zhang, “UA V-Aided Positioning Systems for Ground Devices: Fundamental Limits and Algorithms,”IEEE Internet of Things Journal, vol. 9, no. 15, pp. 13 470– 13 485, 2022

  4. [4]

    Age of Information Based Scheduling for UA V-Aided Localization and Communication,

    T. Liang, T. Zhang, Q. Wu, W. Liu, D. Li, Z. Xie, D. Li, and Q. Zhang, “Age of Information Based Scheduling for UA V-Aided Localization and Communication,”IEEE Transactions on Wireless Communications, vol. 23, no. 5, pp. 4610–4626, 2024

  5. [5]

    M 2-net: Multitask-learning-based multiband signal recognition network,

    X. Zhang, P. Wang, Y . Ma, J. Jiao, S. Wu, and Q. Zhang, “M 2-net: Multitask-learning-based multiband signal recognition network,”IEEE Internet of Things Journal, vol. 12, no. 11, pp. 16 543–16 558, Jun. 2025

  6. [6]

    Space- Based Multi-Dimensional Spectrum Situation Awareness: A Robust Streaming Tensor Subspace Tracking Approach,

    Y . Ma, R. Xiao, X. Zhang, S. Zhang, Y . Gao, and W. Zhang, “Space- Based Multi-Dimensional Spectrum Situation Awareness: A Robust Streaming Tensor Subspace Tracking Approach,”IEEE Transactions on Mobile Computing, vol. 25, no. 4, pp. 4602–4617, Apr. 2026

  7. [7]

    UA V Aided Vehicle Positioning with Imperfect Data Association,

    T. Liang, J. Yang, and T. Zhang, “UA V Aided Vehicle Positioning with Imperfect Data Association,” in2021 IEEE 93rd Vehicular Technology Conference (VTC2021-Spring), 2021, pp. 1–6. IEEE TRANSACTIONS ON NETWORK SCIENCE AND ENGINEERING 14

  8. [8]

    Simultaneous Localization and Tracking for UA V-Enhanced Positioning Network,

    T. Liang and T. Zhang, “Simultaneous Localization and Tracking for UA V-Enhanced Positioning Network,” in2023 IEEE 98th Vehicular Technology Conference (VTC2023-Fall), 2023, pp. 1–5

Show all 90 references
  1. [9]

    Joint Frame Structure and Beamwidth Optimization for Integrated Localiza- tion and Communication,

    T. Liang, Z. Yu, T. Zhang, S. Zhou, W. Liu, D. Li, and Z. Niu, “Joint Frame Structure and Beamwidth Optimization for Integrated Localiza- tion and Communication,” in2024 IEEE Wireless Communications and Networking Conference (WCNC), 2024, pp. 1–6

  2. [10]

    Cooperative Relative Localization for UA V Swarm in GNSS-Denied Environments,

    G. Lei, Y . Ping, T. Liang, H. Ding, and T. Zhang, “Cooperative Relative Localization for UA V Swarm in GNSS-Denied Environments,” inGLOBECOM 2025 - 2025 IEEE Global Communications Conference, 2025, pp. 2264–2269

  3. [11]

    UA V Control and Communication Enabled Low- Altitude Economy: Challenges, Resilient Architecture, and Co-Design Strategies,

    T. Liang, N. Su, Y . Ping, G. Lei, X. Chen, L. Zhou, T. Zhang, Q. Zhang, and T. Q. Quek, “UA V Control and Communication Enabled Low- Altitude Economy: Challenges, Resilient Architecture, and Co-Design Strategies,”arXiv preprint arXiv:2604.04044, 2026

  4. [12]

    Age of information based scheduling for uav aided emergency communication networks,

    T. Liang, W. Liu, J. Yang, and T. Zhang, “Age of information based scheduling for uav aided emergency communication networks,” inICC 2022 - IEEE International Conference on Communications, 2022, pp. 5128–5133

  5. [13]

    Joint Com- munication Topology Formation and Task Offloading for Heterogeneous UA V Swarms,

    G. Lei, T. Liang, H. Ding, Y . Ping, T. Zhang, and Q. Zhang, “Joint Com- munication Topology Formation and Task Offloading for Heterogeneous UA V Swarms,”arXiv preprint arXiv:2604.04045, 2026

  6. [14]

    Toward Seamless Localization and Communication: A Satellite-UA V NTN Architecture,

    T. Liang, T. Zhang, and Q. Zhang, “Toward Seamless Localization and Communication: A Satellite-UA V NTN Architecture,”IEEE Network, vol. 38, no. 4, pp. 103–110, 2024

  7. [15]

    Sensing, Communication, and Control Co-Design for Energy-Efficient UA V-Aided Data Collec- tion,

    T. Liang, T. Zhang, B. Cao, and Q. Zhang, “Sensing, Communication, and Control Co-Design for Energy-Efficient UA V-Aided Data Collec- tion,”IEEE Wireless Communications Letters, vol. 13, no. 10, pp. 2852– 2856, 2024

  8. [16]

    Coop- erative ISAC-Empowered Low-Altitude Economy,

    J. Tang, Y . Yu, C. Pan, H. Ren, D. Wang, J. Wang, and X. You, “Coop- erative ISAC-Empowered Low-Altitude Economy,”IEEE Transactions on Wireless Communications, vol. 24, no. 5, pp. 3837–3853, 2025

  9. [17]

    Distributed Allocation and Scheduling of Tasks with Cross- Schedule Dependencies for Heterogeneous Multi-Robot Teams,

    B. A. Ferreira, T. Petrovic, M. Orsag, J. R. Martinez-de Dios, and S. Bogdan, “Distributed Allocation and Scheduling of Tasks with Cross- Schedule Dependencies for Heterogeneous Multi-Robot Teams,”IEEE Access, vol. 12, pp. 74 327–74 342, 2024

  10. [18]

    Multi-Agent Collaborative Optimization of UA V Trajectory and Latency-Aware DAG Task Offloading in UA V-Assisted MEC,

    C. Zheng, K. Pan, J. Dong, L. Chen, Q. Guo, S. Wu, H. Luo, and X. Zhang, “Multi-Agent Collaborative Optimization of UA V Trajectory and Latency-Aware DAG Task Offloading in UA V-Assisted MEC,”IEEE Access, vol. 12, pp. 42 521–42 534, 2024

  11. [19]

    Towards Secure Low- Altitude Airspace: MLLM-Enabled UA V Intent Recognition,

    G. Lei, T. Liang, Y . Ping, X. Chen, L. Zhou, J. Wu, X. Zhang, H. Ding, X. Zhang, W. Yuan, T. Zhang, and Q. Zhang, “Towards Secure Low- Altitude Airspace: MLLM-Enabled UA V Intent Recognition,”IEEE Internet of Things Magazine, pp. 1–8, 2026

  12. [20]

    Generative AI- Based Dependency-Aware Task Offloading and Resource Allocation for UA V-Assisted IoV,

    X. Wang, C. He, W. Jiang, W. Wang, and X. Liu, “Generative AI- Based Dependency-Aware Task Offloading and Resource Allocation for UA V-Assisted IoV,”IEEE Open Journal of the Communications Society, vol. 6, pp. 3932–3949, 2025

  13. [21]

    Multi-UA V Cooperative Task Offloading and Resource Allocation in 5G Advanced and Beyond,

    H. Guo, Y . Wang, J. Liu, and C. Liu, “Multi-UA V Cooperative Task Offloading and Resource Allocation in 5G Advanced and Beyond,”IEEE Transactions on Wireless Communications, vol. 23, no. 1, pp. 347–359, 2024

  14. [22]

    Joint Task Offloading, Resource Allocation, and Trajectory Design for Multi-UA V Cooperative Edge Computing with Task Priority,

    H. Hao, C. Xu, W. Zhang, S. Yang, and G.-M. Muntean, “Joint Task Offloading, Resource Allocation, and Trajectory Design for Multi-UA V Cooperative Edge Computing with Task Priority,”IEEE Transactions on Mobile Computing, vol. 23, no. 9, pp. 8649–8663, 2024

  15. [23]

    Deep Reinforcement Learning-Based Joint Resource Allocation and Trajectory Design for UA V-Assisted Multi-Cell ISAC Systems,

    H. Zeng, X. Lu, Y . Huang, S. Huang, S. Feng, X. Zhu, and J. Cao, “Deep Reinforcement Learning-Based Joint Resource Allocation and Trajectory Design for UA V-Assisted Multi-Cell ISAC Systems,”IEEE Transactions on Network Science and Engineering, vol. 13, pp. 6383–6401, 2026

  16. [24]

    Communication and control co-design for distributed uav formation,

    Y . Ping, T. Liang, H. Ding, and T. Zhang, “Communication and control co-design for distributed uav formation,” inGLOBECOM 2024 - 2024 IEEE Global Communications Conference, 2024, pp. 5006–5011

  17. [25]

    UA V-Aided Localization and Communication: Joint Frame Structure, Beamwidth, and Power Allocation,

    T. Liang, T. Zhang, S. Zhou, W. Liu, D. Li, and Q. Zhang, “UA V-Aided Localization and Communication: Joint Frame Structure, Beamwidth, and Power Allocation,”IEEE Journal of Selected Areas in Sensors, vol. 1, pp. 154–165, 2024

  18. [26]

    Multi-UA V Collaborative Edge Computing Algorithm for Joint Task Offloading and Channel Resource Allocation,

    Y . Wei, S. Wu, Z. Ji, Z. Yu, C. Jiang, and L. Kuang, “Multi-UA V Collaborative Edge Computing Algorithm for Joint Task Offloading and Channel Resource Allocation,”Journal of Communications and Information Networks, vol. 9, no. 2, pp. 137–150, 2024

  19. [27]

    Unmanned Aerial Vehicle Swarm-Enabled Edge Computing: Potentials, Promising Technologies, and Challenges,

    W. Wu, F. Zhou, B. Wang, Q. Wu, C. Dong, and R. Q. Hu, “Unmanned Aerial Vehicle Swarm-Enabled Edge Computing: Potentials, Promising Technologies, and Challenges,”IEEE Wireless Communications, vol. 29, no. 4, pp. 78–85, 2022

  20. [28]

    Multimodal Large Lan- guage Models-Enabled UA V Swarm: Towards Efficient and Intelligent Autonomous Aerial Systems,

    Y . Ping, T. Liang, H. Ding, G. Lei, J. Wu, X. Zou, K. Shi, R. Shao, C. Zhang, W. Zhang, W. Yuan, and T. Zhang, “Multimodal Large Lan- guage Models-Enabled UA V Swarm: Towards Efficient and Intelligent Autonomous Aerial Systems,”IEEE Wireless Communications, vol. 33, no. 1, pp...

  21. [29]

    Air- Ground Collaboration for Language-Specified Missions in Unknown Environments,

    F. Cladera, Z. Ravichandran, J. Hughes, V . Murali, C. Nieto-Granda, M. Ani Hsieh, G. J. Pappas, C. J. Taylor, and V . Kumar, “Air- Ground Collaboration for Language-Specified Missions in Unknown Environments,”IEEE Transactions on Field Robotics, pp. 1–1, 2025

  22. [30]

    CoT-TL: Low-Resource Temporal Knowledge Representation of Planning Instructions Using Chain-of-Thought Reasoning,

    K. Manas, S. Zwicklbauer, and A. Paschke, “CoT-TL: Low-Resource Temporal Knowledge Representation of Planning Instructions Using Chain-of-Thought Reasoning,” in2024 IEEE/RSJ International Confer- ence on Intelligent Robots and Systems (IROS), 2024, pp. 9636–9643

  23. [31]

    LLM-Enabled Low-Altitude UA V Natural Language Navigation via Signal Temporal Logic Specification Translation and Repair,

    Y . Ping, H. Ding, T. Liang, L. Zhou, G. Lei, X. Chen, J. Wu, J. Zhou, and T. Zhang, “LLM-Enabled Low-Altitude UA V Natural Language Navigation via Signal Temporal Logic Specification Translation and Repair,”arXiv preprint arXiv:2603.27583, 2026

  24. [32]

    TR2MTL: LLM-Based Framework for Metric Temporal Logic Formalization of Traffic Rules,

    K. Manas, S. Zwicklbauer, and A. Paschke, “TR2MTL: LLM-Based Framework for Metric Temporal Logic Formalization of Traffic Rules,” in2024 IEEE Intelligent Vehicles Symposium (IV), 2024, pp. 1206–1213

  25. [33]

    A Review of Task Al- location Methods for UA Vs,

    G. M. Skaltsis, H. S. Shin, and A. Tsourdos, “A Review of Task Al- location Methods for UA Vs,”Journal of Intelligent & Robotic Systems, vol. 109, no. 4, p. Art. no. 76, 2023

  26. [34]

    Capability- Oriented Decision-Making in Multi-UA V Deployment and Task Allo- cation: A Hierarchical Game-Based Framework,

    X. Hai, Q. Feng, W. Chen, C. Wen, and A. W. H. Khong, “Capability- Oriented Decision-Making in Multi-UA V Deployment and Task Allo- cation: A Hierarchical Game-Based Framework,”IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 55, no. 7, pp. 4562–4574, 2025

  27. [35]

    Multi-UA V Task Assignment in Dynamic Environments: Current Trends and Future Directions,

    S. Alqefari and M. E. B. Menai, “Multi-UA V Task Assignment in Dynamic Environments: Current Trends and Future Directions,”Drones, vol. 9, no. 1, p. Art. no. 75, 2025

  28. [36]

    Multi-UA V Collaborative Dynamic Task Allocation Method Based on ISOM and Attention Mechanism,

    J. Wu, J. Zhang, Y . Sun, X. Li, L. Gao, and G. Han, “Multi-UA V Collaborative Dynamic Task Allocation Method Based on ISOM and Attention Mechanism,”IEEE Transactions on Vehicular Technology, vol. 73, no. 5, pp. 6225–6235, 2024

  29. [37]

    A Review of Multi-UA V Task Allocation Algorithms for a Search and Rescue Scenario,

    S. A. Ghauri, M. Sarfraz, R. A. Qamar, M. F. Sohail, and S. A. Khan, “A Review of Multi-UA V Task Allocation Algorithms for a Search and Rescue Scenario,”Journal of Sensor and Actuator Networks, vol. 13, no. 5, p. Art. no. 47, 2024

  30. [38]

    Collabora- tive Task Allocation and Optimization Solution for Unmanned Aerial Vehicles in Search and Rescue,

    D. Han, H. Jiang, L. Wang, X. Zhu, Y . Chen, and Q. Yu, “Collabora- tive Task Allocation and Optimization Solution for Unmanned Aerial Vehicles in Search and Rescue,”Drones, vol. 8, no. 4, p. Art. no. 138, 2024

  31. [39]

    Enhancing Unmanned Aerial Vehicle Task Assignment with the Adaptive Sampling-Based Task Rationality Review Algorithm,

    C. Sun, Y . Yao, and E. Zheng, “Enhancing Unmanned Aerial Vehicle Task Assignment with the Adaptive Sampling-Based Task Rationality Review Algorithm,”Drones, vol. 8, no. 9, p. Art. no. 422, 2024

  32. [40]

    Learning Improvement Heuristics for Multi- Unmanned Aerial Vehicle Task Allocation,

    B. Fan, Y . Bo, and X. Wu, “Learning Improvement Heuristics for Multi- Unmanned Aerial Vehicle Task Allocation,”Drones, vol. 8, no. 11, p. Art. no. 636, 2024

  33. [41]

    Heuristic Monte Carlo Tree Search-Based Task Allocation for Multiple UA Vs in Reconnaissance and Rescue Missions,

    Y . Yang, Y . Sun, L. He, X. Zhou, and S. Hu, “Heuristic Monte Carlo Tree Search-Based Task Allocation for Multiple UA Vs in Reconnaissance and Rescue Missions,” in2025 44th Chinese Control Conference (CCC). IEEE, 2025, pp. 1964–1969

  34. [42]

    Task Assignment of UA V Swarms Based on Deep Reinforcement Learning,

    B. Liu, S. Wang, Q. Li, X. Zhao, Y . Pan, and C. Wang, “Task Assignment of UA V Swarms Based on Deep Reinforcement Learning,”Drones, vol. 7, no. 5, p. Art. no. 297, 2023

  35. [43]

    Multi-UA V Cooperative Target Assignment Method Based on Reinforcement Learning,

    K. Ding, M. Kuang, H. Shi, and J. Gao, “Multi-UA V Cooperative Target Assignment Method Based on Reinforcement Learning,”Drones, vol. 8, no. 10, p. Art. no. 562, 2024

  36. [44]

    A Two-Stage Distributed Task Assignment Algorithm Based on Contract Net Protocol for Multi-UA V Cooperative Reconnaissance Task Reassignment in Dynamic Environments,

    G. Wang, X. Lv, and X. Yan, “A Two-Stage Distributed Task Assignment Algorithm Based on Contract Net Protocol for Multi-UA V Cooperative Reconnaissance Task Reassignment in Dynamic Environments,”Sen- sors, vol. 23, no. 18, p. Art. no. 7980, 2023

  37. [45]

    A Distributed Task Allocation Method for Multi-UA V Systems in Communication-Constrained Environments,

    S. Yan, J. Feng, and F. Pan, “A Distributed Task Allocation Method for Multi-UA V Systems in Communication-Constrained Environments,” Drones, vol. 8, no. 8, p. Art. no. 342, 2024

  38. [46]

    Distributed Task Allocation for Multiple UA Vs Based on Swarm Benefit Optimization,

    Y . Chen, R. Chen, Y . Huang, Z. Xiong, and J. Li, “Distributed Task Allocation for Multiple UA Vs Based on Swarm Benefit Optimization,” Drones, vol. 8, no. 12, p. Art. no. 766, 2024

  39. [47]

    Integrate Assignment of Multiple Het- erogeneous Unmanned Aerial Vehicles Performing Dynamic Disaster Inspection and Validation Task with Dubins Path,

    W. Wu, J. Xu, and Y . Sun, “Integrate Assignment of Multiple Het- erogeneous Unmanned Aerial Vehicles Performing Dynamic Disaster Inspection and Validation Task with Dubins Path,”IEEE Transactions on Aerospace and Electronic Systems, vol. 59, no. 4, pp. 4018–4032, 2023

  40. [48]

    Dynamic Task Allocation for Heterogeneous Multi-UA Vs in Uncertain Environments Based on 4DI- GWO Algorithm,

    H. Huang, Z. Jiang, T. Yan, and Y . Bai, “Dynamic Task Allocation for Heterogeneous Multi-UA Vs in Uncertain Environments Based on 4DI- GWO Algorithm,”Drones, vol. 8, no. 6, p. Art. no. 236, 2024. IEEE TRANSACTIONS ON NETWORK SCIENCE AND ENGINEERING 15

  41. [49]

    Sequential Task Allocation of More Scalable Artificial Dragonfly Swarms Considering Dubins Trajectory,

    Y . Li, D. Wen, S. Zhang, and L. Li, “Sequential Task Allocation of More Scalable Artificial Dragonfly Swarms Considering Dubins Trajectory,” Drones, vol. 8, no. 10, p. Art. no. 596, 2024

  42. [50]

    Task Allocation Algorithm for Heterogeneous UA V Swarm with Temporal Task Chains,

    H. Liu, Z. Shao, Q. Zhou, J. Tu, and S. Zhu, “Task Allocation Algorithm for Heterogeneous UA V Swarm with Temporal Task Chains,”Drones, vol. 9, no. 8, p. Art. no. 574, 2025

  43. [51]

    A Two-Stage Multi-UA V Task Allocation Approach Based on Graph Theory and a Learning-Inspired Immune Algorithm,

    S. Zhang, C. Hu, D. Zhao, K. Yang, Z. Xu, and M. Li, “A Two-Stage Multi-UA V Task Allocation Approach Based on Graph Theory and a Learning-Inspired Immune Algorithm,”Drones, vol. 9, no. 9, p. Art. no. 599, 2025

  44. [52]

    Adaptive Depth Graph Neural Network-Based Dynamic Task Allocation for UA V-UGVs Under Complex Environments,

    Z. Ma, J. Xiong, H. Gong, and X. Wang, “Adaptive Depth Graph Neural Network-Based Dynamic Task Allocation for UA V-UGVs Under Complex Environments,”IEEE Transactions on Intelligent Vehicles, vol. 10, no. 5, pp. 3573–3586, 2025

  45. [53]

    Heterogeneous Multi-Agent Task Allocation Based on Graph Neural Network Ant Colony Optimization Algorithms,

    Z. Ma and H. Gong, “Heterogeneous Multi-Agent Task Allocation Based on Graph Neural Network Ant Colony Optimization Algorithms,” Intelligence & Robotics, vol. 3, no. 4, pp. 581–595, 2023

  46. [54]

    Het- erogeneous GNN-RL-Based Task Offloading for UA V-Aided Smart Agriculture,

    T. Pamuklu, A. Syed, W. S. Kennedy, and M. Erol-Kantarci, “Het- erogeneous GNN-RL-Based Task Offloading for UA V-Aided Smart Agriculture,”IEEE Networking Letters, vol. 5, no. 4, pp. 213–217, 2023

  47. [55]

    Spatiotemporal Attention- Augmented Inverse Reinforcement Learning for Multi-Agent Task Al- location,

    H. Yin, Z. Yang, L. Zhang, and D. Watzenig, “Spatiotemporal Attention- Augmented Inverse Reinforcement Learning for Multi-Agent Task Al- location,”arXiv preprint arXiv:2504.05045, 2025

  48. [56]

    MAGNNET: Multi-Agent Graph Neural Network-Based Efficient Task Allocation for Autonomous Vehicles with Deep Reinforcement Learning,

    L. Ratnabala, A. Fedoseev, R. Peter, and D. Tsetserukou, “MAGNNET: Multi-Agent Graph Neural Network-Based Efficient Task Allocation for Autonomous Vehicles with Deep Reinforcement Learning,” in2025 IEEE Intelligent Vehicles Symposium (IV). IEEE, 2025, pp. 970–975

  49. [57]

    Graph Neural Network Assisted Spectrum Resource Optimisation for UA V Swarm,

    X. Liao, Y . Wang, X. Zhu, C. Lin, Y . Han, and Y . Li, “Graph Neural Network Assisted Spectrum Resource Optimisation for UA V Swarm,” IET Communications, vol. 19, no. 1, p. Art. no. e70078, 2025

  50. [58]

    Graph Neural Network-Enhanced Multi-Agent Reinforcement Learning for Intelligent UA V Confrontation,

    K. Hu, H. Pan, C. Han, J. Sun, D. An, and S. Li, “Graph Neural Network-Enhanced Multi-Agent Reinforcement Learning for Intelligent UA V Confrontation,”Aerospace, vol. 12, no. 8, p. Art. no. 687, 2025

  51. [59]

    Deep Graph Reinforcement Learning for UA V-Enabled Multi-User Secure Communications,

    X. Tang, K. Zhao, C. Shen, Q. Du, Y . Wang, D. Niyato, and Z. Han, “Deep Graph Reinforcement Learning for UA V-Enabled Multi-User Secure Communications,”IEEE Transactions on Mobile Computing, vol. 24, no. 9, pp. 8780–8793, 2025

  52. [60]

    Multi-Hop Differential Topology-Based Al- gorithms for Resilient Network of UA V Swarm,

    H. Lin and L. Ding, “Multi-Hop Differential Topology-Based Al- gorithms for Resilient Network of UA V Swarm,”arXiv preprint arXiv:2411.11342, 2024

  53. [61]

    STAGE: Spatio-Temporal Attention Graph-Enhanced Policy Optimization for Area Defense in UA V Swarm Confrontations,

    Y . Su, Q. Wu, L. Jiang, T. Tong, and W. Tan, “STAGE: Spatio-Temporal Attention Graph-Enhanced Policy Optimization for Area Defense in UA V Swarm Confrontations,”Journal of King Saud University Com- puter and Information Sciences, vol. 38, no. 5, p. Art. no. 198, 2026

  54. [62]

    Dependency-Aware Task Offloading in Multi-UA V Assisted Collaborative Mobile Edge Computing,

    Z. Zhao, X. Xu, T. Zhang, J. Li, and Y . Liu, “Dependency-Aware Task Offloading in Multi-UA V Assisted Collaborative Mobile Edge Computing,”arXiv preprint arXiv:2510.20149, 2025

  55. [63]

    Dependency-Aware Task Collaborative Offloading and Resource Allocation in UA V-Enabled Edge Computing,

    Z. Huang, Z. Kuang, B. Xu, Y . Bi, and A. Liu, “Dependency-Aware Task Collaborative Offloading and Resource Allocation in UA V-Enabled Edge Computing,”Peer-to-Peer Networking and Applications, vol. 18, p. Art. no. 118, 2025

  56. [64]

    UA V- Assisted Dependency-Aware Computation Offloading in Device-Edge- Cloud Collaborative Computing Based on Improved Actor-Critic DRL,

    L. Zhang, R. Tan, Y . Zhang, J. Peng, J. Liu, and K. Li, “UA V- Assisted Dependency-Aware Computation Offloading in Device-Edge- Cloud Collaborative Computing Based on Improved Actor-Critic DRL,” Journal of Systems Architecture, vol. 154, p. Art. no. 103215, 2024

  57. [65]

    A UA V- Enabled Mobile Edge Computing Paradigm for Dependent Tasks Based on a Computing Power Pool,

    X. Lai, Y . Guo, M. He, H. Yuan, W. Li, and X. Cui, “A UA V- Enabled Mobile Edge Computing Paradigm for Dependent Tasks Based on a Computing Power Pool,”Frontiers of Information Technology & Electronic Engineering, vol. 26, no. 4, pp. 623–638, 2025

  58. [66]

    Dependency-Aware Task Offloading Strategy via Heterogeneous Graph Neural Network and Deep Reinforcement Learning,

    J. Wu, Y . Zou, X. Zhang, J. Liu, W. Sun, and G. Du, “Dependency-Aware Task Offloading Strategy via Heterogeneous Graph Neural Network and Deep Reinforcement Learning,”IEEE Internet of Things Journal, vol. 12, no. 13, pp. 22 915–22 933, 2025

  59. [67]

    Post- Disaster Multi-UA V Task Planning Based on Graph Neural Network Decoupling,

    R. Hong, J. Zhang, M. Gan, Q. Wang, B. Xin, and J. Chen, “Post- Disaster Multi-UA V Task Planning Based on Graph Neural Network Decoupling,”Unmanned Systems, vol. 14, no. 3, pp. 589–604, 2026

  60. [68]

    Digital Twin-Based Task-Driven Resource Management in Intelligent UA V Swarms,

    T. Li, S. Leng, X. Liao, and Y . Zhang, “Digital Twin-Based Task-Driven Resource Management in Intelligent UA V Swarms,”IEEE Transactions on Intelligent Transportation Systems, vol. 26, no. 4, pp. 5467–5480, 2025

  61. [69]

    TPML: Task Planning for Multi-UA V System with Large Language Models,

    J. Cui, G. Liu, H. Wang, Y . Yu, and J. Yang, “TPML: Task Planning for Multi-UA V System with Large Language Models,” in2024 IEEE 18th International Conference on Control & Automation (ICCA), 2024, pp. 886–891

  62. [70]

    Swarm-GPT: Combining Large Language Models with Safe Motion Planning for Robot Choreography Design,

    A. Jiao, T. P. Patel, S. Khurana, A.-M. Korol, L. Brunke, V . K. Adajania, U. Culha, S. Zhou, and A. P. Schoellig, “Swarm-GPT: Combining Large Language Models with Safe Motion Planning for Robot Choreography Design,” Extended Abstract in the 6th Robot Learning Workshop at the ...

  63. [71]

    SwarmGPT: Combining Large Language Models with Safe Motion Planning for Drone Swarm Choreography,

    M. Schuck, D. O. Dahanaggamaarachchi, B. Sprenger, V . Vyas, S. Zhou, and A. P. Schoellig, “SwarmGPT: Combining Large Language Models with Safe Motion Planning for Drone Swarm Choreography,”IEEE Robotics and Automation Letters, vol. 10, no. 11, pp. 12 237–12 244, 2025

  64. [72]

    FlockGPT: Guiding UA V Flocking with Linguistic Orchestration,

    A. Lykov, S. Karaf, M. Martynov, V . Serpiva, A. Fedoseev, M. Ko- nenkov, and D. Tsetserukou, “FlockGPT: Guiding UA V Flocking with Linguistic Orchestration,” inProceedings of the IEEE International Symposium on Mixed and Augmented Reality Adjunct. IEEE, 2024

  65. [73]

    A Prompt-Driven Task Planning Method for Multi- Drones Based on Large Language Model,

    Y . Liu and B. Ou, “A Prompt-Driven Task Planning Method for Multi- Drones Based on Large Language Model,” in2025 44th Chinese Control Conference (CCC). IEEE, 2025, pp. 4871–4876

  66. [74]

    CLIPSwarm: Generating Drone Shows from Text Prompts with Vision-Language Models,

    P. Pueyo, E. Montijano, A. C. Murillo, and M. Schwager, “CLIPSwarm: Generating Drone Shows from Text Prompts with Vision-Language Models,” in2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2024, pp. 11 917–11 923

  67. [75]

    SMART-LLM: Smart Multi-Agent Robot Task Planning Using Large Language Models,

    S. S. Kannan, V . L. N. Venkatesh, and B.-C. Min, “SMART-LLM: Smart Multi-Agent Robot Task Planning Using Large Language Models,” in 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2024, pp. 12 140–12 147

  68. [76]

    LSAI: A Large Small AI Model Codesign Framework for Agentic Robot Scenarios,

    L. Zhou, S. Leng, T. Liang, and J. Yao, “LSAI: A Large Small AI Model Codesign Framework for Agentic Robot Scenarios,” 2026

  69. [77]

    Applying Large Language Model to a Control System for Multi-Robot Task Assignment,

    W. Zhao, L. Li, H. Zhan, Y . Wang, and Y . Fu, “Applying Large Language Model to a Control System for Multi-Robot Task Assignment,”Drones, vol. 8, no. 12, p. Art. no. 728, 2024

  70. [78]

    Multi-Robot Task Planning for Multi-Object Retrieval Tasks with Distributed on-Site Knowledge via Large Language Models,

    K. Murata, S. Hasegawa, T. Ishikawa, Y . Hagiwara, A. Taniguchi, L. E. Hafi, and T. Taniguchi, “Multi-Robot Task Planning for Multi-Object Retrieval Tasks with Distributed on-Site Knowledge via Large Language Models,”Artificial Life and Robotics, vol. 31, pp. 418–432, 2026

  71. [79]

    Automatic MILP Model Construction for Multi-Robot Task Allocation and Scheduling Based on Large Language Models,

    M. Peng, Z. Chen, J. Yang, J. Huang, Z. Shi, Q. Liu, X. Li, and L. Gao, “Automatic MILP Model Construction for Multi-Robot Task Allocation and Scheduling Based on Large Language Models,” in2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2...

  72. [80]

    Large Language Models for Multi- Robot Systems: A Survey,

    P. Li, Z. An, S. Abrar, and L. Zhou, “Large Language Models for Multi- Robot Systems: A Survey,”Autonomous Robots, vol. 50, p. Art. no. 30, 2026

  73. [81]

    LiP-LLM: Integrating Linear Programming and Dependency Graph With Large Language Models for Multi-Robot Task Planning,

    K. Obata, T. Aoki, T. Horii, T. Taniguchi, and T. Nagai, “LiP-LLM: Integrating Linear Programming and Dependency Graph With Large Language Models for Multi-Robot Task Planning,”IEEE Robotics and Automation Letters, vol. 10, no. 2, pp. 1122–1129, 2025

  74. [82]

    Demo Abstract: Embodied Aerial Agent for City-Level Visual Language Navigation Using Large Language Model,

    W. Zhang, Y . Liu, X. Wang, X. Chen, C. Gao, and X. Chen, “Demo Abstract: Embodied Aerial Agent for City-Level Visual Language Navigation Using Large Language Model,” in2024 23rd ACM/IEEE International Conference on Information Processing in Sensor Networks (IPSN). IEEE, 2024,...

  75. [83]

    Vision-Language Navigation for Quadcopters with Conditional Transformer and Prompt- Based Text Rephraser,

    Z. Chen, J. Li, F. Fukumoto, P. Liu, and Y . Suzuki, “Vision-Language Navigation for Quadcopters with Conditional Transformer and Prompt- Based Text Rephraser,” inACM Multimedia Asia 2023. ACM, 2023, pp. 1–7

  76. [84]

    NavAgent: Multi-Scale Urban Street View Fusion for UA V Embodied Vision-and- Language Navigation,

    Y . Liu, F. Yao, Y . Yue, G. Xu, X. Sun, and K. Fu, “NavAgent: Multi-Scale Urban Street View Fusion for UA V Embodied Vision-and- Language Navigation,”arXiv preprint arXiv:2411.08579, 2024

  77. [85]

    UA V-VLN: End-to-End Vision-Language Guided Navigation for UA Vs,

    P. Saxena, N. Raghuvanshi, and N. Goveas, “UA V-VLN: End-to-End Vision-Language Guided Navigation for UA Vs,” in2025 European Conference on Mobile Robots (ECMR). IEEE, 2025, pp. 1–6

  78. [86]

    CityNavAgent: Aerial Vision-and-Language Navigation with Hierarchical Semantic Planning and Global Memory,

    W. Zhang, C. Gao, S. Yu, R. Peng, B. Zhao, Q. Zhang, J. Cui, X. Chen, and Y . Li, “CityNavAgent: Aerial Vision-and-Language Navigation with Hierarchical Semantic Planning and Global Memory,” inProceedings of the 63rd Annual Meeting of the Association for Computational Linguist...

  79. [87]

    UA Vs Meet LLMs: Overviews and Perspectives towards Agentic Low-Altitude Mobility,

    Y . Tian, F. Lin, Y . Li, T. Zhang, Q. Zhang, X. Fu, J. Huang, X. Dai, Y . Wang, C. Tian, B. Li, Y . Lv, L. Kov ´acs, and F.-Y . Wang, “UA Vs Meet LLMs: Overviews and Perspectives towards Agentic Low-Altitude Mobility,”Information Fusion, vol. 122, p. Art. no. 103158, 2025

  80. [88]

    AerialVLA: A Vision-Language-Action Model for Aerial Navigation with Online Dia- logue,

    J. Chen, H. Li, Z. Tang, X. Li, W. Wu, and S. Liu, “AerialVLA: A Vision-Language-Action Model for Aerial Navigation with Online Dia- logue,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 40, no. 22. AAAI Press, 2026, pp. 18 161–18 169

  81. [89]

    AutoFly: Vision- Language-Action Model for UA V Autonomous Navigation in the Wild,

    X. Sun, W. Si, W. Ni, Y . Li, D. Wu, F. Xie, R. Guan, H.-Y . Xu, H. Ding, Y . Wu, Y . Yue, Y . Huang, and H. Xiong, “AutoFly: Vision- Language-Action Model for UA V Autonomous Navigation in the Wild,” inInternational Conference on Learning Representations (ICLR), 2026. IEEE TR...

  82. [90]

    Vision- Language Navigation for Aerial Robots: Towards the Era of Large Language Models,

    X. Xia, L. Zhou, Y . Tang, X. Zhu, H. Zhu, and W. Yao, “Vision- Language Navigation for Aerial Robots: Towards the Era of Large Language Models,”arXiv preprint arXiv:2604.07705, 2026

Pith tools

Reviewed July 11, 2026 · model on record in the stance chip above.