Pith. sign in

REVIEW 4 major objections 4 minor 21 references

Barriers on the EDGE: A scalable CBF architecture over EDGE for safe aerial-ground multi-agent coordination

T0 review · 4 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read This paper proposes an edge-computing architecture that uses time-varying control barrier functions to keep UAV-UGV teams safe, including landings on moving ground robots.

desk verdict A plausible edge-computing CBF architecture with a real gap in the NID-based UGV safety argument, and no validation yet; not ready for peer review. read the letter →

arxiv 2411.16608 v2 pith:CBAYKN3Y submitted 2024-11-25 cs.RO cs.SYeess.SY

Viswa Narayanan Sankaranarayanan , Achilleas Santi Seisa , Akshit Saradagi , Sumeet Satpute , George Nikolakopoulos This is my paper · ORCID
classification cs.ROcs.SYeess.SY
keywords controlbarrierfunctionsmulti-agentcoordinationaerial-groundrobotsedgecomputingsafelandingquadraticprogrammingnonholonomicscalable
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 proposes a control architecture that lets teams of aerial and ground robots operate together safely, including the hard case where each UAV lands on its own ground robot while the ground robot keeps moving. The safety constraints—collision avoidance between any two agents, landing clearance above the moving platform, and staying inside a task-space box—are encoded as time-varying Control Barrier Functions (CBFs) and enforced by a small quadratic program running onboard each robot. To keep coordination scalable, a centralized edge node called the Watcher picks out only the constraints that matter for each agent at each moment, so the number of communication links grows linearly with fleet size rather than quadratically. The paper presents the architecture and the CBF constructions; it states explicitly that experimental validation and a full theoretical evaluation are deferred to an extended version.

What carries the argument

The load-bearing object is the time-varying Control Barrier Function (CBF), a function $h(x,t)$ whose superlevel set is kept forward invariant by the control input. Three families are used: spherical and circular CBFs such as $h_{aij} = r_{aij}^2 - s_a^2$ for collision avoidance, the landing CBF $h_{lii}$ of Eq. (12), which is time-varying because the UGV position moves, and box CBFs for the workspace boundary. The paper also relies on a near-identity diffeomorphism (Eq. (8)) that turns the nonholonomic unicycle model into a single-integrator on an offset point $\rho_{oi}$, so the same first-order CBF machinery applies to ground robots. All constraints are assembled by the Watcher node into matrices $(A_{ki}, b_{ki})$ and solved as a quadratic program (15) onboard each agent.

What would settle it

Run two UGVs under the proposed controller for a chosen offset $o_i$ and record the minimum distance between their actual bodies; if it ever falls below the safety radius $s_g$ while all offset-point CBF constraints are satisfied, the near-identity equivalence claim would be refuted.

Watch

Extended reading notes

Core claim

The central claim is that safe, scalable coupled aerial-ground multi-robot operation can be achieved by pairing time-varying CBFs with a hybrid edge-computing topology. For landing, the paper introduces the time-varying landing CBF $h_{lii} = r_{zi} - \beta \alpha l_i \exp(-\alpha l_i) - \gamma$, where $l_i$ is the squared horizontal distance between a UAV and its UGV; this creates a downward funnel above the moving landing platform so the UAV is kept above the platform while descending, and forward invariance of the safe set guarantees it never violates the clearance constraint. Collision avoidance between UAVs, between UGVs, and between UAVs and other UGVs is handled by spherical or circular CBFs in relative-distance variables, while box CBFs keep every agent inside the task space. Scalability comes from the centralized Watcher node on an edge cluster, which computes the set of proximal agents for each robot and transmits only the relevant constraint matrices; each robot then runs its own CBF quadratic-program filter locally, so the per-agent computation and the number of communication links remain modest. The paper claims the star topology uses $2N$ links instead of $2N(2N-1)$ for fully connected designs.

Load-bearing premise

The ground-robot safety guarantees rest on the assumption that steering an offset point on the unicycle through the CBF constraints keeps the actual nonholonomic body safe, but the paper gives no bound on how small the offset must be.

Editorial extensions

If this is right

  • If the architecture works as claimed, a fleet of $N$ UAV-UGV pairs needs only $2N$ communication links instead of $2N(2N-1)$, so adding agents does not blow up connectivity.
  • The time-varying landing CBF gives a forward-invariance guarantee that a UAV remains in a safe funnel above its moving landing platform throughout descent, provided the system starts inside the safe set.
  • Distributed local QP filters mean each robot only needs its own state, its goal setpoint, and the constraint matrices from the Watcher, so onboard processing stays low even as fleet size grows.
  • Because constraints are selected by proximity, agents far apart do not impose constraints on each other, which reduces both computation and conservatism in the CBF filter.

Reading between the lines

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

  • The paper does not derive a bound on the offset distance $o_i$ in the near-identity diffeomorphism; a natural next step is to find how small $o_i$ must be for the original unicycle body to inherit the same safety radius, closing the gap between the offset-point guarantee and the actual robot.
  • The Watcher node is still a single point of failure for constraint selection; one could test whether redundant Watcher replicas or a gossip protocol among agents preserve the safety guarantees under node loss.
  • The landing CBF could be extended to platforms with uncertain or nonzero vertical motion, or to multi-UAV landings on a shared platform, by adding a disturbance term to the time-derivative condition.
  • Since the paper states that experimental validation is deferred, a concrete falsifiable check is to run the proposed controller with the Watcher on a small physical fleet and record whether pairwise distances between UAVs, UGVs, and UAV-UGV pairs stay above their respective safety radii throughout landing.
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

4 major / 4 minor

Summary. The paper proposes a hybrid centralized-distributed control architecture for safe coordination of UAV-UGV pairs in a confined space. Safety is enforced through time-varying Control Barrier Functions (CBFs) for aerial-aerial, ground-ground, aerial-ground obstacle, and landing constraints, while a centralized edge-based "Watcher" node selects the relevant constraints for each agent to reduce communication and onboard computation. The paper provides the problem formulation, the CBF candidate functions, the quadratic-program safety filter, and a discussion of the edge architecture, but it does not include simulations, experiments, or proofs of the main safety claims.

Significance. If the safety and scalability claims were fully established, the architecture could be a useful step toward practical multi-agent aerial-ground coordination: the use of first-order CBF filters with an edge-based constraint-selection layer is a reasonable way to balance onboard and centralized computation, and the near-identity diffeomorphism approach for unicycle UGVs is a known technique with practical appeal. However, the current manuscript does not validate the central claims. There are no experimental or simulation results, no explicit proofs that the proposed functions are valid time-varying CBFs, and no quantitative analysis of the scalability benefits. The paper is best viewed as a preliminary architecture description rather than a complete journal contribution in its present form.

major comments (4)
  1. [Section V] The paper explicitly states: "The validation of the proposed controller and theoretical evaluation will be included in the extended version." This is a load-bearing omission: the primary claims of safe and scalable deployment are not supported by any simulation, experiment, or proof in the submitted manuscript. The paper must either include the validation and theoretical evaluation, or be reframed as a preliminary architecture paper with correspondingly reduced claims.
  2. [Section III-B and Eq. (8)-(9)] The UGV safety filter is implemented in the offset space rho_oi, but the aerial-ground obstacle (AGO) constraint in Eq. (4) and Eq. (11) is defined on the actual UGV origin rho_gj. For a UGV j, h_agj depends on rho_gj = rho_oj - o_j(cos(theta_j), sin(theta_j))^T, which is not a function of the offset state x = rho_oi alone. The QP in Eq. (15) is stated with A = dh/dx and f=0, g=I, but the state x for the UGV is never defined. If x = rho_oi, the AGO constraint matrix cannot be formed as written; if x includes theta_j, the first-order dynamics are more complex than the stated single-integrator form. The paper needs to redefine the AGO constraint consistently in offset coordinates and prove equivalence, or derive the correct augmented-state gradients for the QP.
  3. [Section III-B] The claim that "choosing a sufficiently small o_i" lets the same constraints be used in the offset space is not quantified. Forward invariance of a safe set for the offset point rho_oi does not imply that the physical UGV body (which contains the origin rho_gi) avoids collisions, because the origin can swing around the offset point during rotation. Assumption 2 bounds velocities but not the resulting curvature or heading rate, so the origin can leave a neighborhood of the offset point. To support the ground-ground and aerial-ground safety claims, the paper must provide an explicit bound on o_i and a proof that the original non-holonomic dynamics preserve the safe set for the physical UGV body.
  4. [Section IV-A, IV-B, IV-D] The paper repeatedly states that candidate functions "can be shown" to be valid time-varying CBFs (e.g., Eq. (11), Eq. (12), and the box constraints in Eq. (14)), but no class-K_infinity function xi is provided and no verification of condition (10) is carried out for any constraint. This is especially important for the landing CBF h_lii with the moving target (Eq. (12)), whose partial derivative dh/dt in Eq. (13) depends on the UGV velocity; the paper does not show that the resulting inequality is satisfiable for all admissible control magnitudes. Additionally, the QP (15) may have multiple simultaneously active constraints, and the paper does not address feasibility of the bounded control set when several constraints are active. These gaps directly affect the safety guarantee.
minor comments (4)
  1. [Definition 1] The time interval in Definition 1 is written as [0, infinity], but it should be [0, infinity) since infinity is not an element of the time domain.
  2. [Section II-C] The notation for safety radii in Assumption 1 (s_g > s_ag > s_a) is clear, but the physical rationale is stated only in Remark 1. It would help to state the rationale directly in Assumption 1 or immediately after Eq. (5).
  3. [Section IV-C] The lower altitude bound for the UAV is said to be redundant because the AGC constraint (12) keeps the UAV above the landing platform, but this is only active during landing. If the UAV is not in a landing phase, the lower altitude limit may still be needed; the paper should clarify when AGC is active.
  4. [Section IV-A] The phrase "Provided the UAVs and UGVs initiate with a distance between them larger than their respective safety radii, they do not get closer..." is an assertion, not a proof. Since forward invariance is the central safety mechanism, this statement should be backed by the explicit CBF verification requested in the major comments.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the architecture defines CBF-based safety filters rather than deriving predictions from fitted outputs, and the main risks are unproven approximations, not circular reasoning.

full rationale

The paper's contribution is a control architecture, not an empirical prediction. The safety conditions (3)-(7) are design constraints, and the CBF candidates (11), (12), and (14) are constructed so that h >= 0 directly encodes those constraints; this is the standard CBF construction, not a self-referential derivation. No parameter is fitted to data and then called a prediction. The nearest self-citation is [18], used to supply the landing-funnel formula r_{zi} >= beta alpha l_i exp(-alpha l_i) + gamma in Eq. (6), which is transparently extended to moving targets in Eq. (12). The paper does not use [18] as the proof of validity: it asserts that the candidate 'can be proven' via Definition 1 and leaves the proof to an extended version, so the claim does not reduce to the citation. The NID offset-point step (Eq. (8)) and the statement that 'choosing a sufficiently small o_i' lets the same constraints be used in offset space are unsupported and may be a genuine correctness gap for the physical UGV body, but they are omissions or assumptions, not circularity: the paper never claims to have derived that implication from the equations. Similarly, the AGO constraint matrix computation is not spelled out, which is a completeness issue. Because no load-bearing argument reduces by construction to its inputs, the appropriate finding is no significant circularity.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim rests on standard CBF forward-invariance theory plus several design parameters and domain assumptions that are not quantified or verified in this draft. The missing feasibility proof and the unquantified NID offset are the most fragile pieces.

free parameters (5)
  • α (landing CBF horizontal scaling)
    Design parameter in Eqs. (6) and (12); no value or tuning procedure is given in the draft.
  • β (landing CBF vertical scaling)
    Design parameter in Eqs. (6) and (12); no value or tuning procedure is given.
  • γ (landing CBF offset)
    Offset to avoid turbulence from the landing platform; no value or justification is provided.
  • o_i (NID offset distance)
    Positive offset in Eq. (8); chosen 'sufficiently small' without a formal bound or value.
  • Safety radii s_a, s_g, s_ag
    Design parameters in Eqs. (3)-(5); no values are given and they are assumed to satisfy Assumption 1.
assumptions (5)
  • domain assumption sg > sag > sa (Assumption 1)
    Requires landing platforms to be larger than UAV cross-sections; introduced in Section II.
  • domain assumption UGV velocities are smooth and bounded with |v_i| < v̄ and |ω_i| < ω̄, and the ground is flat (Assumption 2)
    Used to justify the NID approximation and the UAV's ability to track a moving UGV; Section II.
  • domain assumption First-order kinematic model p_dot = u for UAVs and the NID single-integrator approximation for UGVs
    Underpins the CBF filter; Section III-B. The NID transformation replaces non-holonomic dynamics with a single-integrator on an offset point.
  • ad hoc to paper Each candidate function h is a valid time-varying CBF, i.e., there exists a class-K∞ ξ satisfying Eq. (10)
    Asserted without proof for the landing CBF and the spherical/box CBFs in Section IV. No explicit ξ is constructed.
  • ad hoc to paper The QP (15) remains feasible under bounded control sets when multiple constraints are active
    No feasibility analysis is provided; the intersection of safe sets may not be forward invariant when constraints conflict, especially for the landing CBF with bounded velocities.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Barriers on the EDGE: A scalable CBF architecture over EDGE for safe aerial-ground multi-agent coordination." pith.science (2026). https://pith.science/paper/CBAYKN3Y

@misc{pith2026241116608,
  author       = {Pith},
  title        = {Pith review of: Barriers on the EDGE: A scalable CBF architecture over EDGE for safe aerial-ground multi-agent coordination},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CBAYKN3Y}},
  note         = {Machine review of arXiv:2411.16608}
}
read the original abstract

In this article, we propose a control architecture for the safe, coordinated operation of a multi-agent system with aerial (UAVs) and ground (UGVs) robots in a confined task space. We consider the case where the aerial and ground operations are coupled, enabled by the capability of the aerial robots to land on moving ground robots. The proposed method uses time-varying Control Barrier Functions (CBFs) to impose safety constraints associated with (i) collision avoidance between agents, (ii) landing of UAVs on mobile UGVs, and (iii) task space restriction. Further, this article addresses the challenge induced by the rapid increase in the number of CBF constraints with the increasing number of agents through a hybrid centralized-distributed coordination approach that determines the set of CBF constraints that is relevant for every aerial and ground agent at any given time. A centralized node (Watcher), hosted by an edge computing cluster, activates the relevant constraints, thus reducing the network complexity and the need for high onboard processing on the robots. The CBF constraints are enforced in a distributed manner by individual robots that run a nominal controller and safety filter locally to overcome latency and other network nonidealities.

Figures

Figures reproduced from arXiv: 2411.16608 by the authors.

Figure 1
Figure 1. A schematic of multi-robot task scenario. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Edge architecture. Each robot has its own distributed [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 18 canonical work pages

  1. [1]

    Communication and interaction in multi-agent plan- ning,

    M. Georgeff, “Communication and interaction in multi-agent plan- ning,” in Readings in distributed artificial intelligence. Elsevier, 1988, pp. 200–204

    work page 1988

  2. [2]

    Safety-critical containment control for multi-agent systems with communication delays,

    Z. Song, Z. Wu, and H. Huang, “Safety-critical containment control for multi-agent systems with communication delays,” IEEE Transactions on Network Science and Engineering , 2024

    work page 2024

  3. [3]

    Control barrier functions: Theory and applications,

    A. D. Ames, S. Coogan, M. Egerstedt, G. Notomista, K. Sreenath, and P. Tabuada, “Control barrier functions: Theory and applications,” in 2019 18th European Control Conference (ECC) , 2019, pp. 3420– 3431

    2019

  4. [4]

    Advances in the theory of control barrier functions: Addressing practical challenges in safe control synthesis for autonomous and robotic systems,

    K. Garg, J. Usevitch, J. Breeden, M. Black, D. Agrawal, H. Parwana, and D. Panagou, “Advances in the theory of control barrier functions: Addressing practical challenges in safe control synthesis for autonomous and robotic systems,” Annual Reviews in Control , vol. 57, p. 100945, 2024. [Online]. Available: https: //www.sciencedirect.com/science/article/pi...

    work page 2024

  5. [5]

    Safe multi-agent interaction through robust control barrier functions with learned uncertainties,

    R. Cheng, M. J. Khojasteh, A. D. Ames, and J. W. Burdick, “Safe multi-agent interaction through robust control barrier functions with learned uncertainties,” in 2020 59th IEEE Conference on Decision and Control (CDC). IEEE, 2020, pp. 777–783

    work page 2020

  6. [6]

    Control barrier functions for multi-agent systems under conflicting local signal temporal logic tasks,

    L. Lindemann and D. V . Dimarogonas, “Control barrier functions for multi-agent systems under conflicting local signal temporal logic tasks,” IEEE control systems letters , vol. 3, no. 3, pp. 757–762, 2019

    work page 2019

  7. [7]

    Guaranteed obstacle avoidance for multi-robot operations with limited actuation: A control barrier function approach,

    Y . Chen, A. Singletary, and A. D. Ames, “Guaranteed obstacle avoidance for multi-robot operations with limited actuation: A control barrier function approach,” IEEE Control Systems Letters, vol. 5, no. 1, pp. 127–132, 2020

    work page 2020

  8. [8]

    Cbf-inspired weighted buffered voronoi cells for distributed multi-agent collision avoidance,

    Y . Lyu, J. M. Dolan, and W. Luo, “Cbf-inspired weighted buffered voronoi cells for distributed multi-agent collision avoidance,” in 2023 American Control Conference (ACC) . IEEE, 2023, pp. 4513–4518

    work page 2023

Show all 21 references
  1. [9]

    Centralized versus distributed nonlinear model predictive control for online robot fleet trajectory planning,

    F. Bertilsson, M. Gordon, J. Hansson, D. M ¨oller, D. S ¨oderberg, Z. Zhang, and K. ˚Akesson, “Centralized versus distributed nonlinear model predictive control for online robot fleet trajectory planning,” in 2022 IEEE 18th International Conference on Automation Science and En...

    2022

  2. [10]

    Survey of important issues in uav communication networks,

    L. Gupta, R. Jain, and G. Vaszkun, “Survey of important issues in uav communication networks,” IEEE communications surveys & tutorials , vol. 18, no. 2, pp. 1123–1152, 2015

    2015

  3. [11]

    To centralize or not to centralize: A tale of swarm coordination,

    J. Hu, A. Bruno, D. Zagieboylo, M. Zhao, B. Ritchken, B. Jackson, J. Y . Chae, F. Mertil, M. Espinosa, and C. Delimitrou, “To centralize or not to centralize: A tale of swarm coordination,” arXiv preprint arXiv:1805.01786, 2018

    2018 arXiv

  4. [12]

    Distributed implementation of control barrier functions for multi-agent systems,

    X. Tan and D. V . Dimarogonas, “Distributed implementation of control barrier functions for multi-agent systems,” IEEE Control Systems Letters, vol. 6, pp. 1879–1884, 2021

    2021

  5. [13]

    Neural graph control barrier func- tions guided distributed collision-avoidance multi-agent control,

    S. Zhang, K. Garg, and C. Fan, “Neural graph control barrier func- tions guided distributed collision-avoidance multi-agent control,” in Conference on robot learning . PMLR, 2023, pp. 2373–2392

    2023

  6. [14]

    Safety barrier certificates for collisions-free multirobot systems,

    L. Wang, A. D. Ames, and M. Egerstedt, “Safety barrier certificates for collisions-free multirobot systems,” IEEE Transactions on Robotics , vol. 33, no. 3, pp. 661–674, 2017

    2017

  7. [15]

    E-cnmpc: Edge-based centralized nonlinear model predictive control for multiagent robotic systems,

    A. S. Seisa, B. Lindqvist, S. G. Satpute, and G. Nikolakopoulos, “E-cnmpc: Edge-based centralized nonlinear model predictive control for multiagent robotic systems,” IEEE Access , vol. 10, pp. 121 590– 121 601, 2022

    2022

  8. [16]

    Data-driven modeling and distributed predictive control of mixed vehicle platoons,

    J. Zhan, Z. Ma, and L. Zhang, “Data-driven modeling and distributed predictive control of mixed vehicle platoons,” IEEE Transactions on Intelligent Vehicles, 2022

    2022

  9. [17]

    Edge computing architectures for enabling the realisation of the next generation robotic systems,

    A. S. Seisa, G. Damigos, S. G. Satpute, A. Koval, and G. Niko- lakopoulos, “Edge computing architectures for enabling the realisation of the next generation robotic systems,” in 2022 30th Mediterranean conference on control and automation (MED) . IEEE, 2022, pp. 487– 493

    2022

  10. [18]

    A cbf-adaptive control architecture for visual navigation for uav in the presence of uncertainties,

    V . N. Sankaranarayanan, A. Saradagi, S. Satpute, and G. Nikolakopou- los, “A cbf-adaptive control architecture for visual navigation for uav in the presence of uncertainties,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) , 2024, pp. 13 659– 13 665

    2024

  11. [19]

    An edge-based architecture for offloading model predictive control for uavs,

    A. S. Seisa, S. G. Satpute, B. Lindqvist, and G. Nikolakopoulos, “An edge-based architecture for offloading model predictive control for uavs,” Robotics, vol. 11, no. 4, 2022. [Online]. Available: https://www.mdpi.com/2218-6581/11/4/80

    2022

  12. [20]

    The robotarium: Globally impactful opportunities, challenges, and lessons learned in remote-access, distributed control of multirobot systems,

    S. Wilson, P. Glotfelter, L. Wang, S. Mayya, G. Notomista, M. Mote, and M. Egerstedt, “The robotarium: Globally impactful opportunities, challenges, and lessons learned in remote-access, distributed control of multirobot systems,” IEEE Control Systems Magazine , vol. 40, no. 1...

    2020

  13. [21]

    Near-identity diffeomorphisms and exponential/spl epsi/-tracking and/spl epsi/-stabilization of first-order nonholonomic se (2) vehicles,

    R. Olfati-Saber, “Near-identity diffeomorphisms and exponential/spl epsi/-tracking and/spl epsi/-stabilization of first-order nonholonomic se (2) vehicles,” in Proceedings of the 2002 american control conference (ieee cat. no. ch37301) , vol. 6. IEEE, 2002, pp. 4690–4695

    2002

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.