Pith. sign in

REVIEW 3 major objections 5 minor 51 references

Routing stability reduces to keeping canceled requests bounded

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review

2026-08-03 23:26 UTC pith:F5OA5H7M

load-bearing objection Theorem 1 is true but essentially definitional; the paper's real value is the degenerate-stability warning and the observable-cost framing, and the abstract oversells a fleet-sizing extension the body does not make. the 3 major comments →

arxiv 2511.05715 v3 pith:F5OA5H7M submitted 2025-11-07 cs.MA cs.SYeess.SY

Policy Stability for Measuring Operational Performance in Task Assignment with Time-Windows Under Internal Adversarial Influence

Roee M. Francos , Daniel Garces , Orhan Eren Akg\"un , Stephanie Gil This is my paper
classification cs.MA cs.SYeess.SY
keywords average-cost stabilityadversarial location spoofingpickup-and-deliveryrequest time windowsrequest cancellationsfleet sizingdegenerate stabilitymulti-agent task assignment
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper argues that, for pickup-and-delivery fleets under attack by agents that spoof locations and abandon assignments, stability is best judged by observable system signals rather than by route-level work. With bounded arrivals and finite request waiting windows, the number of outstanding requests is automatically capped, so a policy is stable under a long-run average cost if and only if the expected cumulative number of canceled requests stays bounded. The authors extend cooperative fleet-sizing guarantees to time-window settings and show that without time windows, backlog-based stability can certify a badly degraded system as stable, a regime they call degenerate stability. The takeaway is an operational metric: watch the cancellations.

Core claim

The paper's central claim is Theorem 1: under bounded request arrivals and bounded request time windows, the average-cost stability of any assignment policy is equivalent to uniform boundedness of the expected cumulative number of canceled requests. Because each request lives at most w̄ steps and at most η̄ arrive per step, the outstanding count satisfies |R_out| ≤ η̄w̄, and the stage cost g = O_t + C_t then reduces to the question of whether the cumulative cancellation term C_t stays bounded in expectation. This gives stability a direct operational meaning: a fleet is stable exactly when cancellations do not accumulate without bound, regardless of how adversarial agents move or spoof.

What carries the argument

The load-bearing object is the stage cost g(·) = O_t + C_t, the sum of outstanding requests and cumulative canceled requests. The cancellation count C_t is nondecreasing and enters the cost at every step, while the outstanding term is uniformly bounded by η̄w̄ under Assumptions 1 and 2. This asymmetry forces the average cost to be finite only when E[C_t] remains bounded, which is the entire content of the stability characterization.

Load-bearing premise

The main theorem depends on charging the cumulative cancellation count at every time step (g_t = O_t + C_t) with C_t nondecreasing; if the stage cost charged only new cancellations each step, the equivalence between stability and bounded cancellations would no longer hold.

What would settle it

Run a fleet under a policy that cancels requests at a decaying rate, say one extra cancellation every √t steps, with time windows finite. The paper's criterion declares it unstable because the cumulative cost diverges, even though the long-run cancellation rate is zero, demonstrating that the stability verdict rides on the cumulative-charging choice rather than on achievable service quality.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X LinkedIn Reddit HN

If this is right

  • Fleet managers can certify stability from two counters — outstanding and canceled requests — without reconstructing whether agents actually executed assigned routes.
  • A policy that lets requests expire repeatedly is unstable even if the backlog stays flat, closing the loophole of degenerate stability.
  • The same criterion applies unchanged when some agents are adversarial: no route-execution model or worst-case workload loss estimate is needed.
  • Cooperative fleet-sizing bounds carry over to finite time-window settings, giving a baseline for choosing fleet size before adversarial effects are introduced.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The equivalence is driven by the cumulative charging of cancellations in the stage cost; if the cost charged only incremental cancellations, policies with unbounded but sublinear cancellation growth could have finite average cost, so the theorem would fail.
  • A testable extension: monitor the cancellation counter online; a policy whose expected cancellation count breaks its cooperative baseline is a candidate for spoofing, without needing any agent-level trust model.
  • One may reinterpret the result as a Lyapunov-style certificate: because backlog is automatically bounded, the cancellation count is the only state variable that needs to be controlled.
  • The degenerate-stability argument suggests that any backlog-based stability certificate in deadline-constrained systems should be augmented with a wait-time or expiration term.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper studies pickup-and-delivery task assignment with internal adversarial agents that spoof locations to attract and then abandon requests. It proposes an average-cost stability criterion (Eqs. 2–4) whose stage cost is the sum of outstanding requests O_t and cumulative cancellations C_t. Under bounded arrivals and finite request time windows (Assumptions 1–2), Theorem 1 states that finite average cost is equivalent to sup_t E[|R_can^t|] < ∞. The paper also presents feasibility-aware assignment procedures, three adversarial knowledge models, and an empirical study on San Francisco taxi data, arguing that time windows rule out 'degenerate stability' in which bounded backlog hides poor service quality.

Significance. Theorem 1 is correctly proved; under Assumptions 1–2, O_t ≤ η̄w̄ and C_t is nondecreasing, so the equivalence is a standard Cesàro fact for nondecreasing sequences. The paper is useful in proposing a cancellation-aware, observable cost and in empirically comparing it with backlog-only metrics on a realistic dataset. However, the main theoretical result does not use the assignment policy, graph, or adversarial model, and the abstract's claim of extending fleet-sizing guarantees is not supported by any theorem. The contribution is therefore primarily a modeling/empirical one, and the paper should be revised to avoid overclaiming the theoretical content.

major comments (3)
  1. [Section V, Theorem 1 and Eq. (2)] The theorem is correct but largely definitional. From Remark 1, O_t ≤ η̄w̄, and C_t = |R_can^t| is positive and nondecreasing. For any nonnegative nondecreasing sequence a_k, limsup_T (1/T)Σ_{k=t}^T a_k < ∞ iff sup_k a_k < ∞. The proof never uses the assignment policy, graph structure, or adversarial knowledge model. If Eq. (2) were g_t = O_t + ΔC_t with ΔC_t = C_t − C_{t−1}, the equivalence would fail; e.g., E[C_k] = √k gives J_avg = 0 but unbounded cumulative cancellations. Thus Theorem 1 is not an independent operational characterization of policy stability; it is a consequence of charging cumulative cancellations in the stage cost. Please reposition the result as a modeling lemma or add a nontrivial theorem that provides policy-dependent conditions for bounded cancellations.
  2. [Abstract, Sec. VIII-D, Sec. IX] The abstract and introduction claim 'we also extend cooperative fleet-sizing guarantees to finite time-window settings.' No such extension is proved. Section VIII-D only empirically validates the fleet sizes chosen from [21] under expiration, and Section IX lists 'formal fleet-sizing guarantees for time-window-constrained systems' as future work. This claim should be removed or substantiated with a theorem.
  3. [Sec. VIII-F/VIII-G, Figs. 4–5] Because the proposed cost already includes cumulative cancellations, any configuration with a sustained positive cancellation rate will have J_avg = ∞ regardless of routing or adversary details. The empirical growth in Figs. 4 and 5 is therefore a direct consequence of Eq. (2), not an emergent property discovered in simulation. The comparison with the backlog-only metric is illustrative but partly circular. Please state this explicitly and frame the experiments as a demonstration of the cost definition's operational consequences rather than as validation of a nontrivial prediction.
minor comments (5)
  1. [Sec. III-A] Typo: 'Let N denote then total fleet size' should be 'the total fleet size.' Also, Table I uses N for the number of agents without clarifying the relation to the earlier N = |C| + |A|.
  2. [Sec. IV-C, Eq. (3)] The expectation in Eq. (3) is not explicitly conditional on the initial state \x_t; clarify the cost-to-go definition.
  3. [Sec. V-A, proof of Theorem 1] In the forward direction, the argument is correct, but it would be clearer to state that E[C_k] is nondecreasing and therefore sup_k E[C_k] = ∞ implies E[C_k] → ∞.
  4. [Sec. VI-B, Algorithm 1] Algorithm 1 returns only the cooperative assignment vector r_coop^t; the interaction between the adversarial simulation of the auction and the actual server-side assignment is not fully specified.
  5. [Sec. VIII-D] The validation of fleet sizes uses a finite 5,760-step horizon and plateaus, whereas Theorem 1 is an asymptotic statement. This limitation should be stated explicitly.

Circularity Check

1 steps flagged

Theorem 1's equivalence is a Cesàro consequence of the stage cost already charging cumulative cancellations C_t; the central characterization is built into Eq. (2) by construction.

specific steps
  1. self definitional [Sec. IV-C Eq. (2); Sec. V-A Theorem 1 and its proof]
    "We define the stage cost at time t as, g(ˆx_t, µ_t(ˆx_t), η, ρ, δ) = O_t + C_t (2). Theorem 1 (Characterization of average-cost stability): Given Assumptions 1 and 2, J_avg^π(ˆx_t)<∞ for all ˆx_t and all t if and only if sup_{t≥0} E[|R_can^t|]<∞."

    J_avg is the Cesàro mean of E[O_k + C_k]. By Remark 1, O_k ≤ η̄w̄ for every policy, so only E[C_k] matters. C_k is defined as the cumulative number of cancellations and is therefore nondecreasing. For any nondecreasing nonnegative sequence a_k, limsup (1/T)Σ a_k < ∞ iff sup_k a_k < ∞. Thus Theorem 1 is exactly that monotone-sequence lemma applied to the chosen cost; the proof never uses the routing policy, graph structure, or adversarial knowledge model. If Eq. (2) were instead O_t + ΔC_t with incremental cancellations, a policy with E[C_k]=√k would have zero average cost but unbounded sup E[C_k], breaking the announced equivalence. Hence the advertised characterization is baked into the definition of the stage cost, not derived from the model.

full rationale

The paper's headline theoretical contribution reduces to the definition of the stage cost. With g = O_t + C_t, C_t cumulative and nondecreasing, and O_t uniformly bounded by η̄w̄ under Assumptions 1 and 2, finite average cost is equivalent to bounded cumulative cancellations by a standard Cesàro lemma; the supplied proof of Theorem 1 is precisely that argument and uses none of the paper's routing, graph, or adversarial machinery. This is definitional circularity rather than an independent operational characterization. The empirical study and adversarial assignment framework are self-contained and not circular, but the central advertised equivalence is. Separately, I flag that the abstract's claim that the paper extends cooperative fleet-sizing guarantees to finite time-window settings is not supported by any theorem: Sec. IX states that developing formal fleet-sizing guarantees for time-window-constrained systems is future work. That is a missing-support/overclaim issue, not a circularity step, and I did not count it toward the score.

Axiom & Free-Parameter Ledger

2 free parameters · 6 axioms · 0 invented entities

The central theorem rests on two domain assumptions (bounded arrivals, bounded time windows) and the paper's own choice to charge cumulative cancellations in the stage cost; this last choice makes Theorem 1 a near-definitional result. The empirical section additionally uses a hand-picked time window and an empirically fitted greedy fleet size.

free parameters (2)
  • Request time window w_r = D(G) (graph diameter) in main experiments; 0.5D(G) and 2D(G) in validation
    Chosen by hand in Sec. VIII-B/F to keep the cooperative baseline stable while making cancellation meaningful; not derived from theory. The theorem only requires existence of a finite bound w̄.
  • Greedy fleet size |C| = 50
    Empirically calibrated in Sec. VIII-B because no queueing-theoretic guarantee exists for greedy; this fitted value is then used as the cooperative baseline in adversarial experiments.
axioms (6)
  • domain assumption Assumption 1: arrivals are bounded, η_t ≤ η̄ < ∞ for all t.
    Stated in Sec. V-A; used in Remark 1 to bound the backlog. If arrivals are unbounded, |R_out_t| can grow without bound and the equivalence fails.
  • domain assumption Assumption 2: request time windows are uniformly bounded, w_r ≤ w̄ < ∞.
    Stated in Sec. V-A; together with Assumption 1 gives |R_out_t| ≤ η̄w̄. Without finite windows, degeneracy can persist and cancellations vanish.
  • ad hoc to paper Cumulative cancellations are charged at every step: stage cost g_t = O_t + C_t, with C_t nondecreasing (Eq. 2, Def. 4).
    This is the paper's chosen cost model. Theorem 1 is essentially a restatement of this definition plus monotonicity; if per-step cancellations were charged instead, the iff would not hold.
  • standard math Standard real-analysis fact: for a nondecreasing nonnegative sequence a_k, limsup_{T→∞} (1/T) Σ_{k=1}^T a_k < ∞ iff sup_k a_k < ∞.
    Used in the proof of Theorem 1; this is the entire mathematical content beyond the backlog bound.
  • domain assumption Adversarial agents cannot create identities or impersonate other agents; they only spoof reported locations (Def. 1).
    Scopes the attack model for the empirical study; not needed for Theorem 1, but needed for the adversarial framework and simulation claims.
  • domain assumption Request arrivals and pickup/drop-off locations follow a fixed distribution over the horizon (Sec. III-B).
    Used for the cooperative fleet-sizing calibration from [21] and for the simulation. Not needed for Theorem 1.

reviewed 2026-08-03 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Policy Stability for Measuring Operational Performance in Task Assignment with Time-Windows Under Internal Adversarial Influence." pith.science (2026). https://pith.science/paper/F5OA5H7M

@misc{pith2026251105715,
  author       = {Pith},
  title        = {Pith review of: Policy Stability for Measuring Operational Performance in Task Assignment with Time-Windows Under Internal Adversarial Influence},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/F5OA5H7M}},
  note         = {Machine review of arXiv:2511.05715}
}
Share X LinkedIn Reddit HN
read the original abstract

We study autonomous pickup-and-delivery routing problems in which internal adversarial agents spoof their locations to attract request assignments and then intentionally leave those requests unserviced. Such attacks disrupt the centralized scheduler, causing delays, cancellations, and routing instability. A routing policy is stable if its cost remains uniformly bounded over time. Existing policy-cost formulations typically characterize cost through the work required to service outstanding requests. Such a formulation requires analyzing agent-specific route execution and is therefore not well suited to adversarial settings, where non-cooperative agents may arbitrarily deviate from assigned routes or fail to service requests altogether. We introduce a new policy-cost formulation based only on observable system signals, namely the numbers of outstanding and canceled requests. Under bounded arrivals and finite request time windows, we show that stability under this formulation is equivalent to keeping the expected cumulative number of canceled requests uniformly bounded over time, an important operational metric in both cooperative and adversarial settings. We also extend cooperative fleet-sizing guarantees to finite time-window settings and highlight that request time windows are not merely a modeling detail, but are essential for ruling out \emph{degenerate stability}, a regime in which policies are certified as stable despite undesirable large request backlogs.

Figures

Figures reproduced from arXiv: 2511.05715 by Daniel Garces, Orhan Eren Akg\"un, Roee M. Francos, Stephanie Gil.

Figure 1
Figure 1. Figure 1: Investigated policy-knowledge configurations. Rows correspond to assignment policies and columns correspond to adversarial knowledge models, [PITH_FULL_IMAGE:figures/full_fig_p012_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Policy cost for fully cooperative fleets under different request time-window sizes. Results are shown for IA and IA-RA with [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Degenerate stability under the unmonitored spoofing attack models of Def. 1 in the absence of time-window constraints. Since requests do [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Policy cost under adversarial conditions with request time windows equal to one graph diameter. Results are shown for Greedy, IA, and IA-RA [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison between the proposed stability metric and a backlog [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

51 extracted references · 4 linked inside Pith

  1. [1]

    Routing and rebalancing inter- modal autonomous mobility-on-demand systems in mixed traffic,

    S. Wollenstein-Betech, M. Salazar, A. Houshmand, M. Pavone, I. C. Paschalidis, and C. G. Cassandras, “Routing and rebalancing inter- modal autonomous mobility-on-demand systems in mixed traffic,” IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 8, pp. 12 263–12 275, 2021

    2021

  2. [2]

    Group-based distributed auction algorithms for multi-robot task assignment,

    X. Bai, A. Fielbaum, M. Kronm ¨uller, L. Knoedler, and J. Alonso- Mora, “Group-based distributed auction algorithms for multi-robot task assignment,”IEEE Transactions on Automation Science and Engineering, vol. 20, no. 2, pp. 1292–1303, 2022

    2022

  3. [3]

    Graph atten- tion multi-agent fleet autonomy for advanced air mobility,

    M. Fernando, R. Senanayake, H. Choi, and M. Swany, “Graph atten- tion multi-agent fleet autonomy for advanced air mobility,”Robotics: Science and Systems (RSS), 2023

    2023

  4. [4]

    Statistically distinct plans for multi- objective task assignment,

    N. Wilde and J. Alonso-Mora, “Statistically distinct plans for multi- objective task assignment,”IEEE Transactions on Robotics, vol. 40, pp. 2217–2232, 2024

    2024

  5. [5]

    Beyond robustness: A taxonomy of approaches towards resilient multi-robot systems,

    A. Prorok, M. Malencia, L. Carlone, G. S. Sukhatme, B. M. Sadler, and V . Kumar, “Beyond robustness: A taxonomy of approaches towards resilient multi-robot systems,”arXiv:2109.12343, 2021

    Pith/arXiv arXiv 2021

  6. [6]

    Multi-robot coordination and planning in uncertain and adversarial environments,

    L. Zhou and P. Tokekar, “Multi-robot coordination and planning in uncertain and adversarial environments,”Current Robotics Reports, vol. 2, pp. 147–157, 2021

    2021

  7. [7]

    An experimental study of gps spoofing and takeover attacks on uavs,

    H. Sathaye, M. Strohmeier, V . Lenders, and A. Ranganathan, “An experimental study of gps spoofing and takeover attacks on uavs,” in 31st USENIX security symposium (USENIX security 22), 2022, pp. 3503–3520

    2022

  8. [8]

    Unveiling the stealthy threat: Analyzing slow drift gps spoofing attacks for au- tonomous vehicles in urban environments and enabling the resilience,

    S. Dasgupta, A. Ahmed, M. Rahman, and T. N. Bandi, “Unveiling the stealthy threat: Analyzing slow drift gps spoofing attacks for au- tonomous vehicles in urban environments and enabling the resilience,” arXiv:2401.01394, 2024

    Pith/arXiv arXiv 2024

  9. [9]

    Secure trajectory plan- ning against undetectable spoofing attacks,

    Y .-C. Liu, G. Bianchin, and F. Pasqualetti, “Secure trajectory plan- ning against undetectable spoofing attacks,”Automatica, vol. 112, p. 108655, 2020

    2020

  10. [10]

    Multi-agent reinforcement learning-based pas- senger spoofing attack on mobility-as-a-service,

    K.-F. Chu and W. Guo, “Multi-agent reinforcement learning-based pas- senger spoofing attack on mobility-as-a-service,”IEEE Transactions on Dependable and Secure Computing, vol. 21, no. 6, pp. 5565–5581, 2024

    2024

  11. [11]

    Location spoofing attacks on autonomous fleets,

    J. Yang, A. Estornell, and Y . V orobeychik, “Location spoofing attacks on autonomous fleets,” inSymposium on Vehicles Security and Pri- vacy. Internet Society, 2023

    2023

  12. [12]

    Potential cyberattacks on automated vehicles,

    J. Petit and S. E. Shladover, “Potential cyberattacks on automated vehicles,”IEEE Transactions on Intelligent transportation systems, vol. 16, no. 2, pp. 546–556, 2014

    2014

  13. [13]

    A secure sensor fusion framework for connected and automated vehicles under sensor attacks,

    T. Yang and C. Lv, “A secure sensor fusion framework for connected and automated vehicles under sensor attacks,”IEEE Internet of Things Journal, vol. 9, no. 22, pp. 22 357–22 365, 2021

    2021

  14. [14]

    Gps-ids: An anomaly-based gps spoofing attack detection framework for autonomous vehicles,

    M. M. Abrar, A. Youssef, R. Islam, S. Satam, B. S. Latibari, S. Hariri, S. Shao, S. Salehi, and P. Satam, “Gps-ids: An anomaly-based gps spoofing attack detection framework for autonomous vehicles,”arXiv preprint arXiv:2405.08359, 2024

    Pith/arXiv arXiv 2024

  15. [15]

    Secure localization of land vehicles under gps spoofing attack with decomposition kalman filter,

    M. Zhao, Z. Han, E. Wang, and J. Liu, “Secure localization of land vehicles under gps spoofing attack with decomposition kalman filter,” Scientific Reports, 2025

    2025

  16. [16]

    D. P. Bertsekas,Lessons from AlphaZero for optimal, model predictive, and adaptive control. Athena Scientific, 2022

    2022

  17. [17]

    Bertsekas,A course in reinforcement learning

    D. Bertsekas,A course in reinforcement learning. Athena Scientific, 2023

    2023

  18. [18]

    Toward a systematic approach to the design and eval- uation of automated mobility-on-demand systems: A case study in singapore,

    K. Spieser, K. Treleaven, R. Zhang, E. Frazzoli, D. Morton, and M. Pavone, “Toward a systematic approach to the design and eval- uation of automated mobility-on-demand systems: A case study in singapore,”Road Vehicle Automation. Lecture Notes on Mobility, pp. 229–245, 04 2014

    2014

  19. [19]

    Analysis, control, and evaluation of mobility-on-demand systems: A queueing-theoretical approach,

    R. Zhang, F. Rossi, and M. Pavone, “Analysis, control, and evaluation of mobility-on-demand systems: A queueing-theoretical approach,” IEEE Transactions on Control of Network Systems, vol. 6, no. 1, pp. 115–126, 2018

    2018

  20. [20]

    A general maximum-stability dispatch policy for shared autonomous vehicle dispatch with an analytical characterization of the maximum throughput,

    M. W. Levin, “A general maximum-stability dispatch policy for shared autonomous vehicle dispatch with an analytical characterization of the maximum throughput,”Transportation Research Part B: Methodolog- ical, vol. 163, pp. 258–280, 2022

    2022

  21. [21]

    Approximate multiagent reinforcement learning for on-demand urban mobility prob- lem on a large map,

    D. Garces, S. Bhattacharya, D. Bertsekas, and S. Gil, “Approximate multiagent reinforcement learning for on-demand urban mobility prob- lem on a large map,” in2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 6843–6849

    2024

  22. [22]

    Control of robotic mobility-on-demand sys- tems: a queueing-theoretical perspective,

    R. Zhang and M. Pavone, “Control of robotic mobility-on-demand sys- tems: a queueing-theoretical perspective,”The International Journal of Robotics Research, vol. 35, no. 1-3, pp. 186–203, 2016

    2016

  23. [23]

    Craw- dad data set epfl/mobility (v. 2009-02-24),

    M. Piorkowski, N. Sarafijanovic-Djukic, and M. Grossglauser, “Craw- dad data set epfl/mobility (v. 2009-02-24),” 2009

    2009

  24. [24]

    Fms-dispatch: a fast maximum stability dispatch policy for shared autonomous vehicles including exiting passengers under stochastic travel demand,

    T. Xu, M. Cieniawski, and M. W. Levin, “Fms-dispatch: a fast maximum stability dispatch policy for shared autonomous vehicles including exiting passengers under stochastic travel demand,”Trans- portmetrica A: Transport Science, vol. 20, no. 3, p. 2214968, 2024

    2024

  25. [25]

    Smoothing-mp: A novel max- pressure signal control considering signal coordination to smooth traffic in urban networks,

    T. Xu, S. Barman, and M. W. Levin, “Smoothing-mp: A novel max- pressure signal control considering signal coordination to smooth traffic in urban networks,”Transportation Research Part C: Emerging Technologies, vol. 166, p. 104760, 2024

    2024

  26. [26]

    Stable multi-agent routing with bounded-delay adversaries in the decision loop,

    R. M. Francos, D. Garces, and S. Gil, “Stable multi-agent routing with bounded-delay adversaries in the decision loop,” in2025 IEEE 64th Conference on Decision and Control (CDC). IEEE, 2025, pp. 1576–1583

    2025

  27. [27]

    Dynamic games in cyber-physical security: An overview,

    S. R. Etesami and T. Bas ¸ar, “Dynamic games in cyber-physical security: An overview,”Dynamic Games and Applications, vol. 9, no. 4, pp. 884–913, 2019

    2019

  28. [28]

    How physicality enables trust: A new era of trust-centered cyberphysical systems,

    S. Gil, M. Yemini, A. Chorti, A. Nedi ´c, H. V . Poor, and A. J. Goldsmith, “How physicality enables trust: A new era of trust-centered cyberphysical systems,”arXiv preprint arXiv:2311.07492, 2023

    Pith/arXiv arXiv 2023

  29. [29]

    Game-theoretic methods for robustness, security, and resilience of cyberphysical control systems: games-in-games prin- ciple for optimal cross-layer resilient control systems,

    Q. Zhu and T. Basar, “Game-theoretic methods for robustness, security, and resilience of cyberphysical control systems: games-in-games prin- ciple for optimal cross-layer resilient control systems,”IEEE Control Systems Magazine, vol. 35, no. 1, pp. 46–65, 2015

    2015

  30. [30]

    Adversarial reinforce- ment learning for detecting false data injection attacks in vehicular routing,

    T. Eghtesad, Y . V orobeychik, and A. Laszka, “Adversarial reinforce- ment learning for detecting false data injection attacks in vehicular routing,”arXiv preprint arXiv:2603.11433, 2026

    arXiv 2026

  31. [31]

    Sok: Rethinking sensor spoofing attacks against robotic vehicles from a systematic view,

    Y . Xu, X. Han, G. Deng, J. Li, Y . Liu, and T. Zhang, “Sok: Rethinking sensor spoofing attacks against robotic vehicles from a systematic view,” in2023 IEEE 8th European Symposium on Security and Privacy (EuroS&P). IEEE, 2023, pp. 1082–1100

    2023

  32. [32]

    Advgps: Adversarial gps for multi-agent perception attack,

    J. Li, B. Li, X. Liu, J. Fang, F. Juefei-Xu, Q. Guo, and H. Yu, “Advgps: Adversarial gps for multi-agent perception attack,” in2024 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2024, pp. 18 421–18 427

    2024

  33. [33]

    Ghost of the navigator: Spoofing attack against direction of arrival estimation,

    S. Xu, A. Brighente, B. Chen, M. Conti, and S. Peng, “Ghost of the navigator: Spoofing attack against direction of arrival estimation,” IEEE Internet of Things Journal, 2025

    2025

  34. [34]

    Resilient distributed static state estimation for sensor networks against location-varying cyber- attacks,

    X. Lei, G. Wen, D. Zhao, and G. Chen, “Resilient distributed static state estimation for sensor networks against location-varying cyber- attacks,”IEEE Transactions on Automatic Control, 2025

    2025

  35. [35]

    Characterizing trust and resilience in distributed consensus for cyberphysical systems,

    M. Yemini, A. Nedi ´c, A. J. Goldsmith, and S. Gil, “Characterizing trust and resilience in distributed consensus for cyberphysical systems,” IEEE Transactions on Robotics, vol. 38, no. 1, pp. 71–91, 2021

    2021

  36. [36]

    Resilient and efficient multi-robot pickup and delivery against strategic attacks: A three- layered framework,

    X. Gong, H. Li, J. Gui, Z. Shu, and T. Huang, “Resilient and efficient multi-robot pickup and delivery against strategic attacks: A three- layered framework,”IEEE Internet of Things Journal, 2025

    2025

  37. [37]

    Algorithms for the vehicle routing and scheduling problems with time window constraints,

    M. M. Solomon, “Algorithms for the vehicle routing and scheduling problems with time window constraints,”Operations research, vol. 35, no. 2, pp. 254–265, 1987

    1987

  38. [38]

    Vehicle routing problem with time windows, part i: Route construction and local search algorithms,

    O. Br ¨aysy and M. Gendreau, “Vehicle routing problem with time windows, part i: Route construction and local search algorithms,” Transportation science, vol. 39, no. 1, pp. 104–118, 2005

    2005

  39. [39]

    A case study of consistent vehicle routing problem with time windows,

    H. Lespay and K. Suchan, “A case study of consistent vehicle routing problem with time windows,”International Transactions in Operational Research, vol. 28, no. 3, pp. 1135–1163, 2021

    2021

  40. [40]

    The cumulative vehicle routing problem with time windows: models and algorithm,

    A. Fern ´andez Gil, E. Lalla-Ruiz, M. G ´omez S ´anchez, and C. Castro, “The cumulative vehicle routing problem with time windows: models and algorithm,”Annals of Operations Research, vol. 350, no. 1, pp. 325–353, 2025

    2025

  41. [41]

    Dynamic vehicle routing problems,

    H. N. Psaraftis, “Dynamic vehicle routing problems,”Vehicle Routing: Methods and Studies, pp. 223–248, 1988

    1988

  42. [42]

    Dynamic vehicle routing with time windows in theory and practice,

    Z. Yang, J.-P. van Osta, B. van Veen, R. van Krevelen, R. van Klaveren, A. Stam, J. Kok, T. B¨ack, and M. Emmerich, “Dynamic vehicle routing with time windows in theory and practice,”Natural computing, vol. 16, no. 1, pp. 119–134, 2017

    2017

  43. [43]

    Dy- namic vehicle routing: Literature review and modeling framework,

    M. W. Ulmer, J. C. Goodson, D. C. Mattfeld, and B. W. Thomas, “Dy- namic vehicle routing: Literature review and modeling framework,” Transportation Science, vol. 52, no. 1, pp. 255–279, 2017

    2017

  44. [44]

    Recent dynamic vehicle routing problems: A survey,

    B. H. O. Rios, E. C. Xavier, F. K. Miyazawa, P. Amorim, E. Curcio, and M. J. Santos, “Recent dynamic vehicle routing problems: A survey,”Computers & Industrial Engineering, vol. 160, p. 107604, 2021

    2021

  45. [45]

    On-demand high-capacity ride-sharing via dynamic trip-vehicle as- signment,

    J. Alonso-Mora, S. Samaranayake, A. Wallar, E. Frazzoli, and D. Rus, “On-demand high-capacity ride-sharing via dynamic trip-vehicle as- signment,”Proceedings of the National Academy of Sciences, vol. 114, no. 3, pp. 462–467, 2017

    2017

  46. [46]

    Multiagent reinforcement learning for autonomous routing and pickup problem with adaptation to variable demand,

    D. Garces, S. Bhattacharya, S. Gil, and D. Bertsekas, “Multiagent reinforcement learning for autonomous routing and pickup problem with adaptation to variable demand,” in2023 IEEE International Conference on Robotics and Automation (ICRA). IEEE, 2023, pp. 3524–3531

    2023

  47. [47]

    Surge routing: Event-informed multiagent reinforcement learning for autonomous rideshare,

    D. Garces and S. Gil, “Surge routing: Event-informed multiagent reinforcement learning for autonomous rideshare,” inProceedings of the 23rd International Conference on Autonomous Agents and Multiagent Systems, 2024, pp. 641–650

    2024

  48. [48]

    The auction algorithm: A distributed relaxation method for the assignment problem,

    D. Bertsekas, “The auction algorithm: A distributed relaxation method for the assignment problem,”Annals of operations research, vol. 14, no. 1, pp. 105–123, 1988

    1988

  49. [49]

    New auction algorithms for the assignment problem and extensions,

    D. P. Bertsekas, “New auction algorithms for the assignment problem and extensions,”Results in control and optimization, vol. 14, p. 100383, 2024

    2024

  50. [50]

    On implementing 2d rectangular assignment algo- rithms,

    D. F. Crouse, “On implementing 2d rectangular assignment algo- rithms,”IEEE Transactions on Aerospace and Electronic Systems, vol. 52, no. 4, pp. 1679–1696, 2016

    2016

  51. [51]

    D. P. Bertsekas,Rollout, policy iteration, and distributed reinforcement learning. Athena Scientific, 2021

    2021

This paper was first reviewed by deepseek-v4-flash on August 3, 2026.