Pith. sign in

REVIEW 5 major objections 6 minor 2 cited by

SwarmVLM: VLM-Guided Impedance Control for Autonomous Navigation of Heterogeneous Robots in Dynamic Warehousing

T0 review · 5 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read SwarmVLM claims that a drone and ground robot can navigate cluttered, changing warehouse scenes by letting a vision-language model and retrieval-augmented generation choose impedance-control parameters, achieving 92% success over 12 real-wo

desk verdict A credible integration demo whose headline claim — VLM-RAG parameter selection — is never actually tested; the 92% success could be entirely due to the APF+impedance baseline. read the letter →

arxiv 2508.07814 v1 pith:BC7ECWBN submitted 2025-08-11 cs.RO

Malaika Zafar , Roohan Ahmed Khan , Faryal Batool , Yasheerah Yaqoot , Ziang Guo , Mikhail Litvinov , Aleksey Fedoseev , Dzmitry Tsetserukou This is my paper
classification cs.RO
keywords heterogeneousrobotteamsUAV-AGVcoordinationvision-languagemodelretrieval-augmentedgenerationimpedancecontrolartificialpotentialfieldwarehousenavigationdynamicobstacleavoidance
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

The paper sets out to show that a heterogeneous drone–ground-robot team can navigate cluttered, changing indoor spaces without constant human control. Its recipe: the drone plans with an artificial potential field, the ground robot trails it through virtual impedance links (virtual springs and dampers), and a vision-language model reads a top-down camera view to retrieve matching impedance parameters from a six-scenario database. The authors report 11 of 12 real-world trials succeeding (92%) and, in the full text, 80% accuracy for obstacle detection and parameter selection under good lighting; the abstract's matching sentence says 8%, which this extraction treats as a typo. If those numbers hold, the contribution is a working demonstration that vision-language retrieval can set low-level control gains for heterogeneous robot coordination in a warehouse setting.

What carries the argument

The central object is the variable-parameter impedance controller coupled to a retrieval database: Eq. (1) gives the APF forces that guide the drone, Eq. (2) ties the ground robot to the drone as a mass-spring-damper with virtual mass m, damping d, and stiffness k, and Eq. (3) adds a repulsive displacement from short obstacles with coefficient F. The VLM-RAG module selects these four parameters from Table I for six environmental configurations, using sentence-transformer embeddings and an exact nearest-neighbor search in FAISS over 384-dimensional vectors. That six-case parameter table is what carries the claim that adaptive impedance tuning is happening.

What would settle it

Repeat the 12 real-world trials with fixed parameters from a single row of Table I instead of VLM-RAG-retrieved values: if success stays at 11 of 12, the retrieval loop is not doing the claimed work. Conversely, run the same trials in poor lighting, where the paper reports VLM accuracy dropping to 60%; a full-system success rate near 92% would contradict the claim that retrieval accuracy governs navigation.

Watch

Extended reading notes

Core claim

The central claim is that a leader-follower drone–AGV pair can be made adaptive by choosing impedance parameters not by hand per scene but through a VLM-RAG pipeline. The drone carries global navigation with an artificial potential field, while the mobile robot follows through a virtual mass-spring-damper link; when the robot meets short obstacles the drone cannot see, it temporarily breaks the link and applies a repulsive impedance displacement. A ceiling camera feeds the VLM, whose textual description of obstacle number and layout is embedded and matched against a six-case database of 'empirically derived optimal' stiffness, mass, damping, and force coefficients. The paper reports 92% succ

Load-bearing premise

The load-bearing premise is that the six impedance-parameter sets in Table I, found to be optimal in PyBullet simulation, transfer to the real drone and ground robot and remain appropriate when RAG retrieves them in new scenes.

Editorial extensions

If this is right

  • If the 92% success holds, a drone-AGV pair using this architecture can operate in cluttered warehouse aisles without human joystick or gesture control, as long as the scene falls near one of the six database cases.
  • Delegating short-obstacle avoidance to the ground robot lets the drone fly a more direct APF path: in the reported cases the drone's trajectory is about 0.95 to 1.33 m shorter than the robot's.
  • Because the VLM-RAG module sets stiffness, damping, and force coefficients from a top-down image, the team's following behavior can change when obstacles are rearranged, at least within the six-case retrieval set.
  • The reported lateral deviation of up to 50 cm gives a concrete clearance bound for warehouse corridors when the ground robot must dodge low obstacles while keeping formation.

Reading between the lines

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

  • A natural ablation the paper leaves implicit: run the same 12 trials with fixed, non-retrieved impedance parameters; if success stays near 92%, the VLM-RAG retrieval is not actually load-bearing for navigation.
  • The six-case database is likely the scalability bottleneck; an extension would interpolate or regress parameters for unseen obstacle layouts instead of snapping to the nearest stored case.
  • The ceiling-mounted camera restricts the system to instrumented environments; a testable next step is mounting the VLM input on the drone itself to make the team fully self-contained.
  • The abstract's '8%' versus the full text's '80%' in the same sentence needs resolution; the conclusion and Fig. 7 are consistent with 80%, so the lower number is almost certainly a typographical error.
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

5 major / 6 minor

Summary. SwarmVLM proposes a leader-follower heterogeneous system in which a drone plans with an Artificial Potential Field (APF) and a mobile robot follows via virtual impedance links, while a VLM-RAG module (Molmo-7B-D + FAISS) selects impedance parameters from a six-case database. The paper reports a 92% success rate over 12 real-world trials and an 80% VLM-RAG accuracy under good lighting, together with simulation results, trajectory length and deviation tables, velocity profiles, and a demonstration video. The central claim is that the VLM-RAG-generated parameter selection improves navigation in cluttered, dynamic warehousing environments.

Significance. If the VLM-RAG contribution were demonstrated, this would be a useful integration of existing, well-understood components: APF planning, impedance-based formation control, and semantic scene interpretation in a real heterogeneous drone-AGV system. The paper is clearly structured, the real-world feasibility of the baseline controller is plausible, and the reported quantitative information (path offsets, trajectory lengths, dynamic-case behavior) is a strength. However, the current evidence does not establish that the VLM-RAG parameter retrieval affects navigation, the VLM-RAG evaluation is underspecified, and the abstract/body accuracy contradiction (8% vs 80%) leaves the headline claim ill-defined. The paper is therefore not yet ready for publication as a VLM-RAG contribution, though the integration effort itself may be salvageable with additional experiments or a more modest reframing.

major comments (5)
  1. [Abstract and Section V-C] The abstract reports '8% accuracy in object detection and selection of impedance parameters', while Section V-C and the conclusion report '80% success' (also referenced in the reader's strongest claim). This is not a cosmetic inconsistency: the VLM-RAG result is one of the paper's two headline claims, and a reader cannot tell which number is correct. The authors must correct the abstract and state the exact evaluation metric and result.
  2. [Section V-C, Fig. 7] The VLM-RAG evaluation protocol is not defined. There is no statement of the number of test images, the number of runs, the ground-truth labeling procedure, or what counts as a 'success' (object detection IoU? retrieval of the correct case? end-to-end parameter selection?). Lighting conditions are mentioned but not operationalized. Without this information, the reported 80% accuracy cannot be interpreted or reproduced.
  3. [Sections III-C and V-B] No experiment isolates the contribution of the VLM-RAG module to navigation. The 11/12 successes are reported for the full system, but there is no comparison against fixed Table I parameters, random retrieval, or an oracle, and no analysis of which impedance case was retrieved in each trial or whether retrieval errors correlate with the single failure. Since the six parameter sets in Table I are numerically close (m=1.0-1.5, k=3.0-5.0, d=2.5-4.0), the results are consistent with the alternative hypothesis that the APF+impedance baseline alone is robust to this parameter variation and the VLM-RAG retrieval has no measurable effect. This is load-bearing for the paper's stated contribution.
  4. [Section III-C2 and Section V] Table I stores 'empirically derived optimal impedance parameters' obtained from PyBullet simulation using the APF planner, and the real-world trials use these values directly. No sim-to-real transfer analysis, parameter sensitivity study, or stability margin is provided. Because the claimed benefit of VLM-RAG is precisely the selection of these parameters, the paper needs at least a sensitivity or robustness analysis showing that the retrieved values are meaningful beyond the simulation environment in which they were fitted.
  5. [Section V-B] The 92% success rate is based on 12 trials with no confidence interval or failure-detail breakdown; the single failure is attributed to 'synchronization issues' and 'limited flight area due to the large mobile robot's size'. The first attribution is acknowledged, but the paper does not say which cases were tested how many times, whether the failure involved a VLM-RAG retrieval error, or how the 12 trials were distributed among the six database cases. The claim would be strengthened by reporting per-case trial counts and a confidence interval, or by clearly labeling the result as a preliminary feasibility demonstration.
minor comments (6)
  1. [Tables II and III] Table II labels the dynamic case 'IV (Dynamic)', while the text and Table III call it 'CASE III (Dynamic)'. Please align the case numbering consistently.
  2. [Equation (4)] The embedding dimension n=384 is given, but the specific sentence-transformer model is not named. Add the model identifier and a citation so the retrieval pipeline is reproducible.
  3. [Fig. 3 caption] The caption refers to red, black, and blue lines that may be indistinguishable in grayscale print. Consider using distinct line styles or a color-blind-safe palette.
  4. [Abstract and Table II] The abstract mentions a lateral deviation of up to 50 cm, but Table II reports offsets of 0.43-0.45 m. Clarify whether 50 cm corresponds to a trial not in the table or to the maximum across all 12 trials.
  5. [Section III-B, Eq. (3)] For the repulsive displacement Δx_robot = k_impF * r_imp, the units of k_impF and r_imp are not stated. Define all symbols and units, since the equation is central to the local obstacle-avoidance mechanism.
  6. [Section V-A] The text says 12 experiments were conducted with varying obstacle positions, but only Cases I-IV appear in the results. State how the 12 trials map to the six database cases and to the static/dynamic conditions.

Circularity Check

1 steps flagged · score 6.0 of 10

VLM-RAG parameter selection is a retrieval of simulation-fitted database entries; the navigation success is externally measured but not shown to depend on the retrieved parameters.

  1. fitted input called prediction [Section III-C2 (Custom Database) and Section III-C3 (RAG for Impedance Parameter Generation), Table I; evaluated in Section V-C]
    "The database stores empirically derived optimal impedance parameters obtained through simulation experiments using the APF planner for drone navigation. Table I summarizes the parameters for all six cases. ... The closest matching case is returned, and its associated impedance parameters are used to adjust the behavior of the swarm according to the current environmental context."

    The VLM-RAG 'prediction' of impedance parameters is, by construction, one of six vectors fitted in PyBullet simulation with the same APF planner (Table I). Section III-C3 retrieves the nearest stored case, and Section V-C reports 80% success in 'detecting and retrieving' objects against this same six-case database. Thus the reported accuracy validates retrieval consistency against the fitted table, not the correctness of the parameters for navigation. No held-out scene, oracle, or fixed-parameter baseline connects the retrieved values to the 92% real-world success rate, so the parameter-selection result reduces to the database construction.

full rationale

The real-world navigation success (92%) is measured externally through 12 physical trials, so that central feasibility claim is not circular. The APF and impedance equations are standard and cited to external sources (though some citations are self-referential, they are not load-bearing). However, the paper's central novelty—VLM-RAG-guided impedance parameter selection—is circular in a narrower sense: the 'optimal' parameters are empirically fitted in simulation and stored as six cases; the VLM-RAG merely returns the nearest stored case. The 80% accuracy figure is a retrieval/classification score against that same fitted database, not evidence that the retrieved parameters improve navigation. Because no ablation or correlation analysis shows that the VLM-RAG output affects the 11/12 successes, the parameter-selection 'prediction' is forced by construction to reproduce the pre-fitted table. This is a partial circularity: it affects the VLM-RAG contribution, but not the independently measured navigation outcome.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

No new physical entities, forces, mediators, or conserved quantities are introduced; the virtual impedance links and VLM-RAG module are assemblies of existing ideas and components. The central claim rests on several fitted gains and on the assumption that simulation-derived parameters transfer to real hardware.

free parameters (6)
  • APF attraction coefficient k_att
    In Eq. 1, attraction force F_attraction = k_att * d_g; the gain is not reported, so the planner has at least one hand-tuned constant.
  • APF repulsion coefficient k_rep
    In Eq. 1, repulsion force uses k_rep and d_safe; no values are given.
  • APF safe distance d_safe
    Threshold in Eq. 1 at which repulsion activates; not reported.
  • RAG impedance parameter sets (m, k, d, Fcoeff) = Table I: m 1.0-1.5 kg, k 3.0-5.0 N/m, d 2.5-4.0 N.s/m, F 0.15-0.68
    These are the outputs the VLM-RAG 'predicts'; they were empirically fitted in simulation, shown in Table I.
  • Local avoidance gain k_impF and influence radius r_imp
    Eq. 3 computes robot displacement as k_impF * r_imp; neither parameter is identified.
  • PID gains for drone and ground robot
    Section V says custom PID controllers are used, but gains are not given.
assumptions (6)
  • domain assumption Virtual impedance spring-damper (Eq. 2) models drone-robot coupling and remains stable
    Eq. 2 is presented without stability or passivity analysis; the whole formation relies on it.
  • domain assumption Simulation-fitted impedance parameters transfer to real hardware
    Table I values derived in PyBullet are used in real-world trials without re-tuning; Section III-C2 and V.
  • domain assumption VLM-Molmo output can be embedded and matched by FAISS to the right scenario in unseen scenes
    Accuracy is evaluated on the same six-case database; no held-out scenes shown.
  • domain assumption APF global planner reliably avoids tall obstacles
    Eq. 1 is treated as a valid planner, following prior refs [11,24].
  • domain assumption VICON motion capture supplies accurate ground truth
    Real-world tracking depends on VICON per Section V.
  • domain assumption Euclidean distance (Eq. 4) is a suitable embedding similarity for retrieving control parameters
    No analysis of how embedding distance relates to control performance.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SwarmVLM: VLM-Guided Impedance Control for Autonomous Navigation of Heterogeneous Robots in Dynamic Warehousing." pith.science (2026). https://pith.science/paper/BC7ECWBN

@misc{pith2026250807814,
  author       = {Pith},
  title        = {Pith review of: SwarmVLM: VLM-Guided Impedance Control for Autonomous Navigation of Heterogeneous Robots in Dynamic Warehousing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BC7ECWBN}},
  note         = {Machine review of arXiv:2508.07814}
}
read the original abstract

With the growing demand for efficient logistics, unmanned aerial vehicles (UAVs) are increasingly being paired with automated guided vehicles (AGVs). While UAVs offer the ability to navigate through dense environments and varying altitudes, they are limited by battery life, payload capacity, and flight duration, necessitating coordinated ground support. Focusing on heterogeneous navigation, SwarmVLM addresses these limitations by enabling semantic collaboration between UAVs and ground robots through impedance control. The system leverages the Vision Language Model (VLM) and the Retrieval-Augmented Generation (RAG) to adjust impedance control parameters in response to environmental changes. In this framework, the UAV acts as a leader using Artificial Potential Field (APF) planning for real-time navigation, while the ground robot follows via virtual impedance links with adaptive link topology to avoid collisions with short obstacles. The system demonstrated a 92% success rate across 12 real-world trials. Under optimal lighting conditions, the VLM-RAG framework achieved 8% accuracy in object detection and selection of impedance parameters. The mobile robot prioritized short obstacle avoidance, occasionally resulting in a lateral deviation of up to 50 cm from the UAV path, which showcases safe navigation in a cluttered setting.

Figures

Figures reproduced from arXiv: 2508.07814 by the authors.

Figure 1
Figure 1. Framework for adaptive swarm navigation. The system processes [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. System architecture of the proposed SwarmVLM, integrating a VLM-RAG module for impedance estimation with a high-level control system that combines APF-based path planning and impedance control for heterogeneous navigation. the mobile robot is handled via ROS, enabling synchronized behavior through a shared information framework. A. Artificial Potential Fields for Global Path Generation In order for the leader drone … view at source ↗
Figure 3
Figure 3. Experimental setup in the Gym Pybullet environment showing the [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Results in static environment. CASE I: One tall obstacle, one short [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 6
Figure 6. Figure 6: Velocity profile along the trajectory in real world scenarios. [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: Performance of the VLM-RAG system under different lighting [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. A Gesture-Based Visual Learning Model for Acoustophoretic Interactions using a Swarm of AcoustoBots

    cs.RO 2026-04 unverdicted novelty 4.0 of 10

    OpenCLIP-based gesture classification with linear probing controls AcoustoBot swarms at 87.8% accuracy and 3.95 s latency in controlled tests.

  2. Vision-and-Language Navigation for UAVs: Progress, Challenges, and a Research Roadmap

    cs.RO 2026-04 unverdicted novelty 4.0 of 10

    A survey of UAV vision-and-language navigation that establishes a methodological taxonomy, reviews resources and challenges, and proposes a forward-looking research roadmap.

Reference graph

Works this paper leans on

24 extracted references · 20 canonical work pages · cited by 2 Pith papers

  1. [1]

    Inventory robots: Performance eval- uation of an rfid-based navigation strategy,

    S. L ´opez-Soriano and R. Pous, “Inventory robots: Performance eval- uation of an rfid-based navigation strategy,” IEEE Sensors Journal , vol. 23, no. 14, pp. 16 210–16 218, July 2023

    work page 2023

  2. [2]

    Advanced rfid-robot with rotating antennas for smart inventory in high-density shelving systems,

    A. Motroni, S. D’Avella, A. Buffi, P. Tripicchio, M. Unetti, G. Cecchi, and P. Nepa, “Advanced rfid-robot with rotating antennas for smart inventory in high-density shelving systems,” IEEE Journal of Radio Frequency Identification, vol. 8, pp. 559–570, Feb. 2024

    work page 2024

  3. [3]

    Inventory management through mini-drones: Architecture and proof-of-concept implementation,

    D. Cristiani, F. Bottonelli, A. Trotta, and M. Di Felice, “Inventory management through mini-drones: Architecture and proof-of-concept implementation,” in Proc. IEEE Int. Symposium on ”A World of Wireless, Mobile and Multimedia Networks” (WoWMoM) , Aug. 31- Sept. 3, 2020, pp. 317–322

    work page 2020

  4. [4]

    Impedance control: An approach to manipulation,

    N. Hogan, “Impedance control: An approach to manipulation,” in Proc. American Control Conference, June 6-8, 1984, pp. 304–313

    work page 1984

  5. [5]

    Safe operations of an aerial swarm via a cobot human swarm interface,

    S. S. Abdi and D. A. Paley, “Safe operations of an aerial swarm via a cobot human swarm interface,” in Proc. IEEE Int. Conf. on Robotics and Automation (ICRA) , May 29-June 2, 2023, pp. 1701–1707

    work page 2023

  6. [6]

    Intuitive human-swarm interaction with gesture recognition and machine learning,

    G. Khen, D. Zhao, and J. Baca, “Intuitive human-swarm interaction with gesture recognition and machine learning,” in Association for Computing Machinery , New York, NY , USA, Oct. 23-26, 2023, p. 453–456

    work page 2023

  7. [7]

    Path plan- ning with potential field-based obstacle avoidance in a 3d environment by an unmanned aerial vehicle,

    A. Batinovic, J. Goricanec, L. Markovic, and S. Bogdan, “Path plan- ning with potential field-based obstacle avoidance in a 3d environment by an unmanned aerial vehicle,” inProc. IEEE Int. Conf. on Unmanned Aircraft Systems (ICUAS), June 21-24, 2022, pp. 394–401

    work page 2022

  8. [8]

    Adaptive formation control for unmanned aerial vehicles with collision avoidance and switching communication network,

    Y . Yu, C. Chen, J. Guo, M. Chadli, and Z. Xiang, “Adaptive formation control for unmanned aerial vehicles with collision avoidance and switching communication network,” IEEE Transactions on Fuzzy Systems, vol. 32, no. 3, pp. 1435–1445, March 2024

    work page 2024

Show all 24 references
  1. [9]

    SwarmTouch: Guiding a swarm of micro- quadrotors with impedance control using a wearable tactile interface,

    E. Tsykunov, R. Agishev, R. Ibrahimov, L. Labazanova, A. Tleugazy, and D. Tsetserukou, “SwarmTouch: Guiding a swarm of micro- quadrotors with impedance control using a wearable tactile interface,” IEEE Transactions on Haptics, vol. 12, no. 3, pp. 363–374, July 2019

    2019

  2. [10]

    DandelionTouch: High fidelity haptic rendering of soft objects in vr by a swarm of drones,

    A. Fedoseev, A. Baza, A. Gupta, E. Dorzhieva, R. N. Gujarathi, and D. Tsetserukou, “DandelionTouch: High fidelity haptic rendering of soft objects in vr by a swarm of drones,” in Proc. IEEE Int. Conf. on Systems, Man, and Cybernetics (SMC) , Oct. 9-12, 2022, pp. 1078– 1083

    2022

  3. [11]

    SwarmPath: Drone swarm navigation through cluttered environments leveraging artificial potential field and impedance control,

    R. A. Khan, M. Zafar, A. Batool, A. Fedoseev, and D. Tsetserukou, “SwarmPath: Drone swarm navigation through cluttered environments leveraging artificial potential field and impedance control,” in Proc. IEEE Int. Conf. on Robotics and Biomimetics (ROBIO) , Dec. 10-14, 2024, pp...

    2024

  4. [12]

    SwarmGear: Heterogeneous swarm of drones with mor- phogenetic leader drone and virtual impedance links for multi-agent inspection,

    Z. Darush, M. Martynov, A. Fedoseev, A. Shcherbak, and D. Tset- serukou, “SwarmGear: Heterogeneous swarm of drones with mor- phogenetic leader drone and virtual impedance links for multi-agent inspection,” in Proc. IEEE Int. Conf. on Unmanned Aircraft Systems (ICUAS), June 6-9...

    2023

  5. [13]

    MorphoLander: Reinforcement learning based landing of a group of drones on the adaptive morphogenetic uav,

    S. Karaf, A. Fedoseev, M. Martynov, Z. Darush, A. Shcherbak, and D. Tsetserukou, “MorphoLander: Reinforcement learning based landing of a group of drones on the adaptive morphogenetic uav,” in Proc. IEEE Int. Conf. on Systems, Man, and Cybernetics (SMC) , Oct. 1-4, 2023, pp. 2507–2512

    2023

  6. [14]

    Target search and navigation in heterogeneous robot systems with deep reinforcement learning,

    Y . Chen and J. Xiao, “Target search and navigation in heterogeneous robot systems with deep reinforcement learning,” Machine Intelligence Research, vol. 22, no. 1, p. 79–90, Jan 2025

    2025

  7. [15]

    The concept of an autonomous mobile robot for automating transport tasks in high-bay warehouses,

    W. Małopolski and S. Skoczypiec, “The concept of an autonomous mobile robot for automating transport tasks in high-bay warehouses,” Advances in Science and Technology Research Journal , vol. 18, pp. 1–10, April 2024

    2024

  8. [16]

    Collaborative object search using heterogeneous mobile robots,

    W. L. Salas, L. M. Valent ´ın-Coronado, I. Becerra, and A. Ram ´ırez- Pedraza, “Collaborative object search using heterogeneous mobile robots,” in Proc. IEEE Int. Autumn Meeting on Power, Electronics and Computing (ROPEC) , vol. 5, Nov. 10-12, 2021, pp. 1–6

    2021

  9. [17]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner et al., “An image is worth 16x16 words: Transformers for image recognition at scale,” 2021, arXiv:2010.11929

    2021 arXiv

  10. [18]

    RT-1: Robotics transformer for real-world control at scale,

    A. Brohan, N. Brown, J. Carbajal, Y . Chebotar, J. Dabis, C. Finn et al. , “RT-1: Robotics transformer for real-world control at scale,” 2023, arXiv:2212.06817

    2023 arXiv

  11. [19]

    Molmo and PixMo: Open weights and open data for state-of-the-art multimodal models,

    M. Deitke, C. Clark, S. Lee, R. Tripathi, Y . Yang, J. S. Park et al. , “Molmo and PixMo: Open weights and open data for state-of-the-art multimodal models,” 2024, arXiv:2409.17146

    2024 arXiv

  12. [20]

    Flockgpt: Guiding uav flocking with linguistic orchestration,

    A. Lykov, S. Karaf, M. Martynov, V . Serpiva, A. Fedoseev, M. Ko- nenkov, and D. Tsetserukou, “Flockgpt: Guiding uav flocking with linguistic orchestration,” in 2024 IEEE International Symposium on Mixed and Augmented Reality Adjunct (ISMAR-Adjunct) , Oct. 21-15, 2024, pp. 485–488

    2024

  13. [21]

    HetSwarm: Cooperative navigation of heterogeneous swarm in dynamic and dense environments through impedance-based guidance,

    M. Zafar, R. A. Khan, A. Fedoseev, K. K. Jaiswal, P. B. Sujit, and D. Tsetserukou, “HetSwarm: Cooperative navigation of heterogeneous swarm in dynamic and dense environments through impedance-based guidance,” in Proc. IEEE Int. Conf. on Unmanned Aircraft Systems (ICUAS), May 1...

    2025

  14. [22]

    ImpedanceGPT: Vlm-driven impedance control of swarm of mini-drones for intelligent navigation in dynamic environment,

    F. Batool, M. Zafar, Y . Yaqoot, R. A. Khan, M. H. Khan, A. Fe- doseev, and D. Tsetserukou, “ImpedanceGPT: Vlm-driven impedance control of swarm of mini-drones for intelligent navigation in dynamic environment,” 2025, arxiv:2503.02723

    2025

  15. [23]

    Learning to fly—a gym environment with pybullet physics for reinforcement learning of multi-agent quadcopter control,

    J. Panerati, H. Zheng, S. Zhou, J. Xu, A. Prorok, and A. P. Schoellig, “Learning to fly—a gym environment with pybullet physics for reinforcement learning of multi-agent quadcopter control,” in 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 20...

    2021

  16. [24]

    Robotic path planning strategy based on improved artificial potential field,

    H. Li, “Robotic path planning strategy based on improved artificial potential field,” in Proc. Int. Conf. on Artificial Intelligence and Computer Engineering (ICAICE) , Oct. 23-25, 2020, pp. 67–71

    2020

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.