Pith. sign in

REVIEW 5 major objections 5 minor 44 references

Conflict Resolution under Degraded Surveillance in Air Corridors Using Multi-Agent Reinforcement Learning

T0 review · 5 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read Decentralized DQN-based reinforcement learning can resolve separation conflicts among heterogeneous UAVs and eVTOLs in a structured corridor under degraded surveillance, and the learned policies can screen traffic-density and separation con

desk verdict The paper sells a degraded-surveillance MARL result that its own equations do not actually deliver: the policy inputs are ground-truth states, and the action space is described as 14 actions while the network outputs 6. read the letter →

arxiv 2607.20547 v1 pith:6SOB2UTA submitted 2026-07-13 cs.LG

Esrat Farhana Dulia , Syed Arbab Mohd Shihab , Caleb Adams , Ruben Del Rosario This is my paper
classification cs.LG
keywords AdvancedAirMobilitymulti-agentreinforcementlearningdeepQ-networkconflictresolutiondegradedsurveillanceUAVeVTOLcorridor
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 multi-agent deep reinforcement learning approach can resolve mid-air separation conflicts in a structured air corridor even when surveillance information is noisy, delayed, or incomplete. If true, this would give Advanced Air Mobility operators an onboard, decentralized conflict-resolution capability that does not depend on continuous ground-based surveillance. The authors train separate DQN policies for small UAVs and eVTOL aircraft, then evaluate them across 90 combinations of traffic density and separation threshold. The headline findings are that loss-of-separation events become more frequent as density and separation requirements grow, but most are resolved within a second, and that agents mostly use turning and speed control during conflicts rather than vertical maneuvers or landings. The paper also uses Pareto analysis to identify six configurations that balance corridor capacity against safety.

What carries the argument

The core mechanism is the interaction between (1) a 28-dimensional local observation state (self position, velocity, energy, plus relative positions, velocities, and information-age for the three nearest neighbors), (2) a 14-action discrete control space with separate DQN policy networks for UAVs and eVTOLs, and (3) a reward function that rewards destination alignment and lane containment while penalizing loss-of-separation and collisions. The environment injects degradation through noisy position/velocity/energy observations, communication delay, information dropout, and process uncertainty. The two separate DQNs are the load-bearing learning component: they allow each aircraft type to lear

What would settle it

Inspect the simulation code or equations: (1) confirm whether the neighbor set N_i,t uses true positions (Eq. 49) rather than observed positions, and whether neighbor velocities in the state (Eq. 50) are the true velocities or the noisy reconstructed velocities from Eq. (30); (2) verify the DQN output layer: the architecture table lists 6 outputs, but the paper claims a 14-action space. A concrete test: run the trained policy in an environment where neighbor detection itself has missed detections and false alarms (not just 5% dropout), and measure whether LOS rate and resolution time degrade.

Watch

Extended reading notes

Core claim

The central claim is that a DQN-based MARL model with separate policies for UAV and eVTOL agents can perform decentralized conflict resolution in a degraded surveillance environment. The environment models a two-layer corridor with cruise and passing lanes, and injects observation noise, communication delay, information dropout, wind, and actuator/model uncertainty. Each agent observes its own state and up to three nearest neighbors and chooses from a discrete action set that includes maintaining course, turning, vertical maneuvers, landing, and speed changes. Across 90 tested density/separation configurations, collisions were essentially eliminated; LOS events increased with density and sep

Load-bearing premise

The paper's headline result depends on the assumption that the 'degraded surveillance' actually degrades what the agents perceive — but the neighbor set and neighbor velocities in the state are computed from true positions and true velocities, so agents always know which aircraft are nearby and how fast they are moving; only positions (after delay) and self-observations are noisy.

Editorial extensions

If this is right

  • If the policies work as described, AAM aircraft could resolve conflicts with only local onboard information, reducing reliance on ground surveillance.
  • The Pareto front (4 agents at 175 m separation; 16, 18, 20, 24 agents at 150 m) gives concrete capacity-safety trade-off points for corridor design.
  • LOS duration analysis shows most events resolve in 1 s, suggesting that tactical RL policies can respond quickly enough for corridor operations.
  • The absence of landing actions in the results indicates that in-flight maneuvers (turn, speed, vertical) are sufficient for the tested scenarios.
  • Low reward variability across repeated episodes suggests the learned policies are consistent across random initial traffic arrangements.

Reading between the lines

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

  • The framework could be used to set corridor capacity limits: at 150 m minimum separation, up to 24 agents appear feasible, and this type of simulation could guide airspace design.
  • A natural next step is to make the neighbor set itself uncertain (missed detections, false alarms) rather than only perturbing observed positions, since the current model assumes agents always know which neighbors are present.
  • The action distribution during conflicts suggests that speed control is underused relative to turning; retraining with different reward weights could reveal whether speed-based resolution reduces corridor excursions.
  • The 1 s time step and quick resolution times suggest coupling these tactical policies with a strategic path planner, so that long-range conflicts are avoided while RL handles short-range encounters.
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 / 5 minor

Summary. The paper proposes a Deep Q-Network (DQN)-based multi-agent reinforcement learning (MARL) framework for decentralized conflict resolution among heterogeneous small UAVs and eVTOL aircraft operating in a structured three-dimensional air corridor. The environment includes aircraft-specific dynamics, energy consumption, corridor constraints, and a simulated degraded-surveillance model with observation noise, communication delay, information dropout, wind, actuator uncertainty, and model uncertainty. Separate DQN policies are trained for UAV and eVTOL agents using a claimed 14-action space, then evaluated on 90 combinations of traffic density and minimum-separation threshold. The reported results include loss-of-separation frequency and duration, action-selection shares (e.g., 33% turning, 25% speed control during LOS), Pareto-optimal configurations, and stability/convergence checks. The central claim is that the framework supports simulation-based evaluation of conflict-resolution strategies under degraded surveillance.

Significance. If the results were supported as stated, this would be a useful integrative simulation contribution: it combines heterogeneous aircraft, structured corridors, detailed energy models, and multiple observation-degradation mechanisms into a single MARL testbed, and it evaluates policies across a broad scenario grid. The paper also reports reproducibility details (fixed seeds, deterministic cuDNN settings) and separates training from testing. However, the central claim is not currently demonstrated because the state definition feeds the policies ground-truth self and neighbor information, the DQN output dimension does not match the claimed 14-action space, and several quantitative results are internally inconsistent. These are load-bearing issues rather than presentation problems.

major comments (5)
  1. [III.B.1, Eqs. (46)-(50)] The state vector used by the DQN contains ground-truth self states (x,y,z,vx,vy,vz,E) and neighbor states (Δx,Δy,Δz,vx_j,vy_j,vz_j,τ) with no hats, and the neighbor set in Eq. (49) is defined using exact positions (||p_j,t - p_i,t|| <= R_local). The observation-noise model in Eqs. (28)-(33) defines \hat p, \hat v, \hat E, delayed positions, and dropout, but none of these are substituted into S_i,t in Eq. (50). In particular, Eq. (30) reconstructs noisy neighbor velocities, yet Eq. (47) uses raw velocities. Thus the policies are trained and evaluated on exact state information, and the paper's headline claim of conflict resolution under degraded surveillance is not supported. This is the central load-bearing gap; correcting it requires feeding noisy/delayed/dropped observations into the state and rerunning the experiments.
  2. [III.B.2 vs. III.B.5, Table 2 vs. Table 3, Eq. (73)] The action space is specified as 14 actions (A={0,...,13}, Table 2), but the DQN architecture outputs one Q-value for each of six discrete actions, with the output layer given as 6 in Table 3 and Eq. (73) listing Q(S,0) through Q(S,5). The greedy selection in Eq. (74) therefore cannot choose among the 14 actions; speed-control actions, which are actions 7-14 in Table 2, are outside the network output. The reported 25% speed-control share during LOS cannot be produced by the described network. If the 6-output statement is a typo, the mapping from Q-values to the 14 actions must be made explicit. As written, the architecture and the experimental results are irreconcilable.
  3. [VI.B-VI.D, Table 6] There is an inconsistency in the Pareto-optimal configuration set. The text in VI.B says the Pareto-optimal configurations include 4 agents with 175 m separation and higher-density configurations with 16, 18, 20, and 24 agents at 150 m, which is five configurations. Later, the paper says six Pareto-optimal configurations were identified, and Table 6 includes a 6-agent/150 m scenario. This discrepancy affects the selection of scenarios for the action analysis and stability checks. Please clarify the actual Pareto front and ensure the reported results correspond to the listed scenarios.
  4. [VI.C, Figures 5-7] The representative two-agent encounter shows initial speeds of approximately 34 m/s and later speeds around 28.7-33.2 m/s. According to Eq. (12) and Eq. (13), the maximum possible initial speed is 26×1.10 = 28.6 m/s for eVTOL and 7×1.10 = 7.7 m/s for UAV. The plotted speeds cannot be produced by the described dynamics. Please reconcile these values, explain if the figures use a different speed model or units, or remove this example as unsupported.
  5. [VI.D, Eq. (70)] The 'model validation' relies on self-defined thresholds: a relative change of 2.76% in the 100-episode moving-average reward at two adjacent episodes is called convergence because it is below a 3% threshold, and testing stability uses a coefficient of variation below 10%. These thresholds are arbitrary, and the testing is performed in the same simulator used for training with reward weights hand-chosen in Eq. (70). The validation therefore establishes internal consistency but not that the policies generalize to different degradation levels or that the learned behavior reflects true robustness to degraded surveillance. The conclusion should be scoped much more cautiously, and ideally the paper should compare against baselines or ablation variants.
minor comments (5)
  1. [Table 2 / Eq. (52)] The action set is written as A={0,...,13} in Eq. (52), but Table 2 labels actions 1 through 14 and does not define action 0. Please make the indexing consistent throughout.
  2. [I.A] There is a missing reference in the phrase 'concepts of operation [?]' in the first paragraph of the introduction.
  3. [III.A.2 / Eq. (62)] During training, the minimum-separation thresholds are sampled separately for eVTOL and UAV, but Eq. (57) uses the threshold of aircraft i only for a pairwise indicator. The treatment of heterogeneous separation thresholds in a pair should be stated explicitly.
  4. [III.B.5] The notation for Q-values in Eqs. (71)-(73) is confusing: Eq. (73) writes q as a vector of six entries but then the greedy action in Eq. (74) is taken over A with 14 actions. This should be rewritten consistently once the action/output mismatch is resolved.
  5. [VI.A / Table 5] The paper states that 'no emergency landing or controlled landing action was selected' in any scenario. This is surprising given that the action space includes landing actions; if landing actions are available, their absence should be briefly explained (e.g., due to reward shaping or rare necessity).

Circularity Check

1 steps flagged · score 2.0 of 10

No significant circularity: the DQN training and held-out evaluation are self-contained; the only by-construction element is the auxiliary 'LOS increases with separation threshold' trend, which follows from Eq. (57) since the threshold is absent from the policy state. The degradation-in-inputs gap (Eqs. 46-50 vs. 28-33) is a correctness issue, not circularity.

  1. self definitional [Abstract; Section III.B.3 Eq. (57); Section III.B.1 Eqs. (46)-(50); Section VI.A]
    "Eq. (57): L_{ij,t}=1 when d_{ij,t}<D^{min}, else 0. Sec. VI.A: "LOS events also increased as the minimum-separation threshold became larger because a wider safety buffer caused more aircraft pairs to violate the required separation distance.""

    The LOS indicator is defined in Eq. (57) as a threshold on pairwise distance. The threshold Dmin appears nowhere in the policy state S_{i,t} (Eq. 50) and test initialization (Sec. III.B.8) is independent of s, so for fixed density the frozen policies yield statistically identical trajectories at every s in {150,...,350}. The LOS count is then exactly the number of pairs with d_{ij,t}<s - the empirical pairwise-distance CDF - which is monotonically non-decreasing by definition of Eq. (57). The headline 'LOS frequency increases with separation requirements' is thus a restatement of the metric's definition applied to unchanged trajectories, not an emergent property of the learned policies. The action-share and LOS-duration results are not definitional; only this threshold trend reduces by con

full rationale

The central derivation is self-contained: two DQN policies are trained with temporal-difference updates, experience replay, and target networks (Eqs. 71-88) on randomized traffic scenarios, then frozen and evaluated on an independently generated 90-scenario grid (Section III.B.8). The headline statistics - LOS counts/durations, SAFE/LOS action shares (79% maintain; during LOS: 33% turning, 29% maintain, 25% speed, 13% vertical), the two-agent encounter - are emergent simulation outputs, not parameters fitted to a target; the fitted-input-called-prediction pattern does not occur. Reward weights (Eq. 70) and the validation thresholds (2.76% vs. 3%; CV < 10%) are author-chosen, which weakens external validity but is not circularity: the reported quantities are not predictions of the fitted weights. Self-citations ([1], [3], [7]-[11], [15], [16]) support only background claims (AAM economics, surveillance sensing, degraded-surveillance motivation), usually alongside external references; no load-bearing modeling choice, uniqueness theorem, or ansatz is imported from them. The one definitional element is flagged as a step: because the LOS indicator (Eq. 57) is a distance-threshold test and the threshold is absent from the state (Eq. 50) and scenario initialization, the 'LOS increases with separation requirement' trend is a monotonicity property of Eq. (57) over statistically identical trajectories - a low-severity, auxiliary result. The paper's genuine weaknesses are correctness/support gaps, per its own text: (i) the central 'degraded surveillance' claim is unsupported because Eqs. (46)-(50) feed the DQN ground-truth self/neighbor positions and velocities; the hatted quantities of Eqs. (28)-(33), the delay model (Eq. 31), and P_drop=0.05 are never substituted into S_{i,t}; (ii) the 14-action space (Eq. 52) contradicts the 6-output DQN (Table 3, Eq. 73); (iii) the paper concedes validation 'does not imply certification-level proof of real-world safety' (Sec. VI.D) and results 'do not provide certification-level evidence of real-world safety' (Sec. VII). None of these reduces a result to its inputs, so they are weighed in this verdict rather than counted as circularity steps.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The framework rests on a large set of author-chosen constants, reward weights, and model choices, none calibrated to flight data. The free parameters below are the ones that most directly determine the reported numbers.

free parameters (7)
  • Reward weights (1.5, 0.2, -1.0, 10, 30) = Eq. (70)
    Hand-chosen shaping terms; the reported action distribution and LOS frequency depend directly on these relative weights.
  • Noise standard deviations sigma_w, sigma_a, sigma_m, sigma_p,self
    Defined in Eqs. (26)-(29) but numeric values are never reported; the degraded-surveillance regime is therefore unspecified.
  • Dropout probability P_drop = 0.05
    Chosen in Eq. (33) without calibration; controls how often neighbor data disappear.
  • Communication delay bound tau_max = 0.5 s (uniform)
    Delay model in Eq. (31) is an unvalidated choice.
  • Stability/convergence thresholds = 3% and CV < 10%
    Thresholds used in Section VI.D to declare convergence and stability; chosen post hoc without independent justification.
  • Power-model constants (W, n, A, delta, s, C_T, k, v0, S_fp, etc.)
    Needed for Eqs. (35)-(44) but not provided; energy behavior is not reproducible.
  • Neighborhood size K=3 and radius R_local=1500 m = 3, 1500 m
    Observation architecture choices; R_local equals corridor length, so almost all traffic is visible unless dropped.
assumptions (5)
  • domain assumption Degraded surveillance is adequately represented by additive Gaussian noise, uniform delay, Bernoulli dropout, and wind/actuator/model noise (Eqs. 26-33).
    Stated as assumptions in Section IV; no sensor model or empirical calibration.
  • domain assumption Point-mass kinematic dynamics with Delta_t = 1 s capture conflict-relevant behavior.
    Section III.A.3 uses simplified kinematics; no six-degree-of-freedom validation.
  • domain assumption The hand-crafted reward function (Eq. 70) represents safe and efficient operation.
    Section IV explicitly says the reward function was assumed to represent objectives.
  • ad hoc to paper Agents can determine the neighbor set using true positions (Eq. 49).
    This contradicts the degraded-observation premise; it is not declared as an assumption.
  • standard math Standard DQN convergence applies despite multi-agent nonstationarity.
    Independent DQN updates with replay buffer are used; the literature review acknowledges nonstationarity but the method does not address it.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Conflict Resolution under Degraded Surveillance in Air Corridors Using Multi-Agent Reinforcement Learning." pith.science (2026). https://pith.science/paper/6SOB2UTA

@misc{pith2026260720547,
  author       = {Pith},
  title        = {Pith review of: Conflict Resolution under Degraded Surveillance in Air Corridors Using Multi-Agent Reinforcement Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6SOB2UTA}},
  note         = {Machine review of arXiv:2607.20547}
}
read the original abstract

Safe Advanced Air Mobility operations require aircraft to maintain separation when surveillance information is noisy, delayed, incomplete, or temporarily unavailable. This study develops a Deep Q-Network-based Multi-Agent Reinforcement Learning framework for decentralized conflict resolution among heterogeneous small unmanned aerial vehicles and electric vertical takeoff and landing aircraft operating within a structured three-dimensional corridor. Separate policies are trained for the two aircraft categories using local observations and a 14-action space that includes maintaining course, turning, vertical maneuvering, landing, and speed control. The simulation incorporates aircraft-specific dynamics, energy use, corridor constraints, observation noise, communication delay, information dropout, wind disturbance, actuator uncertainty, and model uncertainty. The trained policies are evaluated across 90 combinations of traffic density and minimum separation thresholds. Loss-of-separation frequency and duration generally increase with traffic density and separation requirements, although most events are resolved within 1s. Under safe conditions, agents maintain their motion approximately 79% of the time. During conflicts, turning accounts for 33% of actions, followed by maintaining motion at 29%, speed control at 25%, and vertical maneuvers at 13%. Six Pareto-optimal configurations reveal trade-offs between safety and corridor capacity. The framework supports the simulation-based evaluation of safer AAM conflict-resolution strategies under degraded surveillance conditions.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

44 extracted references · 10 canonical work pages

  1. [1]

    Benefits of Advanced Air Mobility for Society and Environment: A Case Study of Ohio,

    Dulia, E. F., Sabuj, M. S., and Shihab, S. A. M., “Benefits of Advanced Air Mobility for Society and Environment: A Case Study of Ohio,”Applied Sciences, Vol. 12, No. 1, 2021, p. 207

    2021

  2. [2]

    Infrastructure To Support Advanced Autonomous Aircraft Technologies in Ohio: Economic Impact Report for Advanced Autonomous Aircraft Technologies in Ohio,

    Rosario, R. D., Davis, T., Dyment, M., and Cohen, K., “Infrastructure To Support Advanced Autonomous Aircraft Technologies in Ohio: Economic Impact Report for Advanced Autonomous Aircraft Technologies in Ohio,” Tech. rep., Ohio Department of Transportation, Office of Statewide Planning and Research, Columbus, OH, 2021

    2021

  3. [3]

    Open Framework Standards for Combined Aircraft Sensor Network for the State of Ohio to Detect and Track Lower Altitude Aircraft: Cost-Benefit Analysis,

    Dulia, Esrat Farhana and Shihab, Syed A. M., “Open Framework Standards for Combined Aircraft Sensor Network for the State of Ohio to Detect and Track Lower Altitude Aircraft: Cost-Benefit Analysis,” Tech. Rep. Agreement No.: 36496, PID: 114242, SJN: 136337, U.S. Department of Transportation, December 2023. https://doi.org/10.13140/RG.2.2.17752.92166

    arXiv 2023

  4. [4]

    Open Framework Standards for Combined Aircraft Sensor Network for the State of Ohio to Detect and Track Lower Altitude Aircraft,

    Calhoun, S., Adami, T., Lorenzon, J., Shihab, S., Bair, J., Wagner, E., Tanner, B., Reck, N., Runneals, D., and McCrink, M., “Open Framework Standards for Combined Aircraft Sensor Network for the State of Ohio to Detect and Track Lower Altitude Aircraft,” Tech. Rep. FHWA/OH-2023-17, Ohio Department of Transportation, Office of Statewide Planning and Resea...

    2023

  5. [5]

    Concept of Operations v2.0, Unmanned aircraft System (UAS) Traffic Management (UTM),

    Federal Aviation Administration (FAA), “Concept of Operations v2.0, Unmanned aircraft System (UAS) Traffic Management (UTM),” https://www.faa.gov/uas/research_development/traffic_management/media/UTM_ConOps_v2.pdf, accessed August 16, 2022

    2022

  6. [6]

    Urban air mobility (UAM) market study,

    Hasan, S., “Urban air mobility (UAM) market study,” 2018

    2018

  7. [7]

    Building Public–Private Partnerships for Advanced Air Mobility Infrastructure Using Game Theory,

    Dulia, E. F., and Shihab, S. A., “Building Public–Private Partnerships for Advanced Air Mobility Infrastructure Using Game Theory,”Journal of Air Transportation, 2024, pp. 1–17. https://doi.org/https://doi.org/10.2514/1.D0447

    work page doi:10.2514/1.d0447 2024

  8. [8]

    How to Negotiate with Private Investors for Advanced Air Mobility Infrastructure? An Analysis of Public Private Partnerships Using Game Theory,

    Dulia, E. F., and Shihab, S. A. M., “How to Negotiate with Private Investors for Advanced Air Mobility Infrastructure? An Analysis of Public Private Partnerships Using Game Theory,”AIAA AVIATION FORUM AND ASCEND 2024, 2024, p. 3782

    2024

Show all 44 references
  1. [9]

    Low-Noise Route and Operations Planning for Food Delivery UnmannedAerialVehicles,

    Dulia, E. F., Shihab, S. A. M., Adams, C., and Burnett, J., “Low-Noise Route and Operations Planning for Food Delivery UnmannedAerialVehicles,”AIAAAVIATIONFORUMANDASCEND2025,2025,p.3303.https://doi.org/10.2514/6.2025-3303

    2025 doi

  2. [10]

    Dynamic Dispatching of Package Delivery UAVs Considering Safety Risk and Noise and Visual Pollution,

    Dulia, E. F., Adams, C. W., and Shihab, S. A. M., “Dynamic Dispatching of Package Delivery UAVs Considering Safety Risk and Noise and Visual Pollution,” , 2025. https://doi.org/10.13140/RG.2.2.18444.83842/2

    2025

  3. [11]

    An integrated supply chain network design for advanced air mobility aircraft manufacturing using stochastic optimization,

    Dulia, E. F., and Shihab, S. A., “An integrated supply chain network design for advanced air mobility aircraft manufacturing using stochastic optimization,”Supply Chain Analytics, Vol. 8, 2024, p. 100083. https://doi.org/10.1016/j.sca.2024.100083, URL https://doi.org/10.1016/j...

    2024

  4. [12]

    Urban Air Mobility Concept of Operations, Version 2.0,

    Federal Aviation Administration, “Urban Air Mobility Concept of Operations, Version 2.0,” Tech. rep., Federal Aviation Ad- ministration, 2023. URL https://www.faa.gov/sites/faa.gov/files/Urban-Air-Mobility-Concept-of-Operations-2.0.pdf, accessed: 2026-06-13

    2023

  5. [13]

    Reliable, Secure, and Scalable Communications,Navigation,andSurveillanceOptionsforUrbanAirMobility,

    Stouffer, V. L., Cotton, W. B., DeAngelis, R. A., Devasirvatham, D. M. J., Irvine, T. B., et al., “Reliable, Secure, and Scalable Communications,Navigation,andSurveillanceOptionsforUrbanAirMobility,”Tech.Rep.NASA/CR–2020-5006661,National Aeronautics and Space Administration, 2...

    2020

  6. [14]

    Next-Generation Airborne Collision Avoidance System,

    Kochenderfer, M. J., Holland, J. E., and Chryssanthacopoulos, J. P., “Next-Generation Airborne Collision Avoidance System,” Lincoln Laboratory Journal, Vol. 19, No. 1, 2012, pp. 17–33. URL https://www.ll.mit.edu/r-d/publications/next-generation- airborne-collision-avoidance-sy...

    2012

  7. [15]

    Designing a Surveillance Sensor Network with Information Clearinghouse for Advanced Air Mobility,

    Dulia, E. F., and Shihab, S. A. M., “Designing a Surveillance Sensor Network with Information Clearinghouse for Advanced Air Mobility,”Sensors, Vol. 24, No. 3, 2024, p. 803. https://doi.org/10.3390/s24030803

    2024 doi

  8. [16]

    Reliability, Robustness, and Resilience Modeling for Surveillance Systems in Advanced Air Mobility Operations,

    Dulia, E. F., Adams, C., Shihab, S. A. M., and Del Rosario, R., “Reliability, Robustness, and Resilience Modeling for Surveillance Systems in Advanced Air Mobility Operations,”AIAA Aviation Forum and ASCEND 2026, 2026. Accepted

    2026

  9. [17]

    A Survey of Recent Results in Networked Control Systems,

    Hespanha, J. P., Naghshtabrizi, P., and Xu, Y., “A Survey of Recent Results in Networked Control Systems,”Proceedings of the IEEE, Vol. 95, No. 1, 2007, pp. 138–162. https://doi.org/10.1109/JPROC.2006.887288, URL https://doi.org/10.1109/JPROC. 2006.887288

    2007

  10. [18]

    Kalman Filtering with Intermittent Observations,

    Sinopoli, B., Schenato, L., Franceschetti, M., Poolla, K., Jordan, M. I., and Sastry, S. S., “Kalman Filtering with Intermittent Observations,”IEEE Transactions on Automatic Control, Vol. 49, No. 9, 2004, pp. 1453–1464. https://doi.org/10.1109/TAC. 2004.834121, URL https://doi...

    2004

  11. [19]

    Robust Deep Reinforcement Learning against Adversarial Perturbations on State Observations,

    Zhang, H., Chen, H., Xiao, C., Li, B., Liu, M., Boning, D., and Hsieh, C.-J., “Robust Deep Reinforcement Learning against Adversarial Perturbations on State Observations,”Advances in Neural Information Processing Systems, Vol. 33, 2020, pp. 21024–21037. URL https://proceedings...

    2020

  12. [20]

    Reinforcement Learning with Delayed Observations,

    Agarwal, M., and Aggarwal, V., “Reinforcement Learning with Delayed Observations,”Proceedings of the International ConferenceonAutomatedPlanningandScheduling, Vol.31, 2021, pp.2–10. URLhttps://ojs.aaai.org/index.php/ICAPS/article/ view/15940, accessed: 2026-06-13

    2021

  13. [21]

    Urban Air Mobility Airspace Integration Concepts and Considerations,

    Thipphavong, D. P., Apaza, R., Barmore, B., Battiste, V., Belcastro, C., Burian, B., Dao, Q., Feary, M., Go, S., Goodrich, K. H., Homola, J., Idris, H., Kopardekar, P., Lachter, J., Neogi, N., Ng, H. K., Oseguera-Lohr, R., Patterson, M. D., and Verma, S. A., “Urban Air Mobilit...

    2018 doi

  14. [22]

    Designing Airspace for Urban Air Mobility: A Review of Concepts and Approaches,

    Bauranov, A., and Rakas, J., “Designing Airspace for Urban Air Mobility: A Review of Concepts and Approaches,” Progress in Aerospace Sciences, Vol. 125, 2021, p. 100726. https://doi.org/10.1016/j.paerosci.2021.100726, URL https: //doi.org/10.1016/j.paerosci.2021.100726

    2021

  15. [23]

    A Review of Conflict Detection and Resolution Modeling Methods,

    Kuchar, J. K., and Yang, L. C., “A Review of Conflict Detection and Resolution Modeling Methods,”IEEE Transactions on Intelligent Transportation Systems, Vol. 1, No. 4, 2000, pp. 179–189. https://doi.org/10.1109/6979.898217, URL https://doi.org/10.1109/6979.898217

  16. [24]

    A Multi-Agent Reinforcement Learning Approach for Conflict Resolution in Dense Traffic Scenarios,

    Lai, J., Cai, K.-Q., Liu, J., and Yang, X., “A Multi-Agent Reinforcement Learning Approach for Conflict Resolution in Dense Traffic Scenarios,”2021 IEEE/AIAA 40th Digital Avionics Systems Conference, IEEE, 2021. https://doi.org/10.1109/ DASC52595.2021.9594437

    2021

  17. [25]

    Autonomous Separation Assurance with Deep Multi-Agent Reinforcement Learning,

    Brittain, M. W., Yang, X., and Wei, P., “Autonomous Separation Assurance with Deep Multi-Agent Reinforcement Learning,” Journal of Aerospace Information Systems, Vol. 18, No. 12, 2021, pp. 890–905. https://doi.org/10.2514/1.I010973, URL https://doi.org/10.2514/1.I010973

    2021 doi

  18. [26]

    Motion Planning in Dynamic Environments Using Velocity Obstacles,

    Fiorini, P., and Shiller, Z., “Motion Planning in Dynamic Environments Using Velocity Obstacles,”The International Journal of Robotics Research, Vol. 17, No. 7, 1998, pp. 760–772. https://doi.org/10.1177/027836499801700706, URL https://doi.org/10.1177/027836499801700706

    1998 doi

  19. [27]

    Conflict-FreeFour-DimensionalPathPlanningforUrbanAirMobilityConsideringAirspace Occupancy,

    Dai,W.,Pang,B.,andLow,K.H.,“Conflict-FreeFour-DimensionalPathPlanningforUrbanAirMobilityConsideringAirspace Occupancy,”Aerospace Science and Technology, Vol. 119, 2021, p. 107154. https://doi.org/10.1016/j.ast.2021.107154. 44

    2021

  20. [28]

    Automated Flight Planning of High-Density Urban Air Mobility,

    Tang, H., Zhang, Y., Mohmoodian, V., and Charkhgard, H., “Automated Flight Planning of High-Density Urban Air Mobility,” Transportation Research Part C: Emerging Technologies, Vol. 131, 2021, p. 103324. https://doi.org/10.1016/j.trc.2021.103324

    2021

  21. [29]

    Human-Level Control through Deep Reinforcement Learning,

    Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., Graves, A., Riedmiller, M., Fidjeland, A. K., Ostrovski, G., Petersen, S., Beattie, C., Sadik, A., Antonoglou, I., King, H., Kumaran, D., Wierstra, D., Legg, S., and Hassabis, D., “Human-Level Co...

    2015 doi

  22. [30]

    Multi-AgentActor-CriticforMixedCooperative-Competitive Environments,

    Lowe,R.,Wu,Y.,Tamar,A.,Harb,J.,Abbeel,P.,andMordatch,I.,“Multi-AgentActor-CriticforMixedCooperative-Competitive Environments,”Advances in Neural Information Processing Systems, Vol. 30, 2017. URL https://arxiv.org/abs/1706.02275, accessed: 2026-06-13

    2017 arXiv

  23. [31]

    One to Any: Distributed Conflict Resolution with Deep Multi-Agent Reinforcement Learning,

    Brittain, M. W., Yang, X., and Wei, P., “One to Any: Distributed Conflict Resolution with Deep Multi-Agent Reinforcement Learning,”AIAA AVIATION 2021 Forum, American Institute of Aeronautics and Astronautics, 2021. https://doi.org/10.2514/6. 2021-1952

    2021 doi

  24. [32]

    Multi-UAV Conflict Resolution with Graph Convolutional Reinforcement Learning,

    Isufaj, R., Omeri, M., and Piera, M. A., “Multi-UAV Conflict Resolution with Graph Convolutional Reinforcement Learning,” AppliedSciences,Vol.12,No.2,2022,p.610. https://doi.org/10.3390/app12020610,URLhttps://doi.org/10.3390/app12020610

    2022 doi

  25. [33]

    Strategic Conflict Management Using Recurrent Multi-Agent Reinforcement Learning for Urban Air Mobility Operations Considering Uncertainties,

    Huang, C., Petrunin, I., and Tsourdos, A., “Strategic Conflict Management Using Recurrent Multi-Agent Reinforcement Learning for Urban Air Mobility Operations Considering Uncertainties,”Journal of Intelligent & Robotic Systems, Vol. 108, No. 1, 2023, p. 8. https://doi.org/10.1...

    2023 doi

  26. [34]

    Autonomous Conflict Resolution in Urban Air Mobility: A Deep Multi-Agent Reinforcement Learning Approach,

    Deniz, S., and Wang, Z., “Autonomous Conflict Resolution in Urban Air Mobility: A Deep Multi-Agent Reinforcement Learning Approach,”AIAA AVIATION Forum and ASCEND, 2024. https://doi.org/10.2514/6.2024-4005, URL https: //doi.org/10.2514/6.2024-4005

    2024 doi

  27. [35]

    AReinforcementLearningApproachtoVehicleCoordinationforStructuredAdvanced Air Mobility,

    Deniz,S.,Wu,Y.,Shi,Y.,andWang,Z.,“AReinforcementLearningApproachtoVehicleCoordinationforStructuredAdvanced Air Mobility,”Green Energy and Intelligent Transportation, Vol. 3, No. 2, 2024, p. 100157. https://doi.org/10.1016/j.geits.2024. 100157

    2024 doi

  28. [36]

    Self-Organized Free-Flight Arrival for Urban Air Mobility,

    Waltz, M., Okhrin, O., and Schultz, M., “Self-Organized Free-Flight Arrival for Urban Air Mobility,”Transportation Research Part C: Emerging Technologies, Vol. 167, 2024, p. 104806. https://doi.org/10.1016/j.trc.2024.104806

    2024

  29. [37]

    Constrained Urban Airspace Design for Large-Scale Drone-Based Delivery Traffic,

    Doole, M., Ellerbroek, J., Knoop, V. L., and Hoekstra, J. M., “Constrained Urban Airspace Design for Large-Scale Drone-Based Delivery Traffic,”Aerospace, Vol. 8, No. 2, 2021, p. 38. https://doi.org/10.3390/aerospace8020038, URL https://doi.org/10.3390/aerospace8020038

    2021 doi

  30. [38]

    Air Traffic Assignment for Intensive Urban Air Mobility Operations,

    Wang, Z., Delahaye, D., Farges, J.-L., and Alam, S., “Air Traffic Assignment for Intensive Urban Air Mobility Operations,” Journal of Aerospace Information Systems, Vol. 18, No. 11, 2021, pp. 860–875. https://doi.org/10.2514/1.I010954

    2021 doi

  31. [39]

    Safe and Scalable Real-Time Trajectory Planning Framework for Urban Air Mobility,

    Taye, A. G., Valenti, R., Rajhans, A., Mavrommati, A., Mosterman, P. J., and Wei, P., “Safe and Scalable Real-Time Trajectory Planning Framework for Urban Air Mobility,”Journal of Aerospace Information Systems, Vol. 21, No. 8, 2024, pp. 641–650. https://doi.org/10.2514/1.I011381

    2024 doi

  32. [40]

    Reinforcement Learning with Random Delays,

    Ramstedt, S., Bouteiller, Y., Beltrame, G., Pal, C., and Binas, J., “Reinforcement Learning with Random Delays,”International Conference on Learning Representations, 2021. URL https://openreview.net/forum?id=QFYnKlBJYR, accessed: 2026-06-13

    2021

  33. [41]

    Probabilistic Modeling and Reasoning of Conflict Detection Effectiveness by Tracking Systems Towards Safe Urban Air Mobility Operations,

    Dai, W., Quek, Z. H., and Low, K. H., “Probabilistic Modeling and Reasoning of Conflict Detection Effectiveness by Tracking Systems Towards Safe Urban Air Mobility Operations,”Reliability Engineering & System Safety, Vol. 244, 2024, p. 109908. https://doi.org/10.1016/j.ress.20...

    2024

  34. [42]

    Modeling Power Consumptions for Multirotor UAVs,

    Gong, H., Huang, B., Jia, B., and Dai, H., “Modeling Power Consumptions for Multirotor UAVs,”IEEE Transactions on Aerospace and Electronic Systems, Vol. 59, No. 6, 2023, pp. 7409–7422. https://doi.org/10.1109/TAES.2023.3288846, URL https://doi.org/10.1109/TAES.2023.3288846

    2023

  35. [43]

    Data-Efficient Modeling for Precise Power Consumption Estimation of Quadrotor Operations Using Ensemble Learning,

    Dai, W., Zhang, M., and Low, K. H., “Data-Efficient Modeling for Precise Power Consumption Estimation of Quadrotor Operations Using Ensemble Learning,” , 2022. URL https://arxiv.org/abs/2205.10997, accessed: 2026-06-13

    2022 arXiv

  36. [44]

    Energy-Efficient Arrival with RTA Constraint for Multirotor eVTOL in Urban Air Mobility,

    Pradeep, P., and Wei, P., “Energy-Efficient Arrival with RTA Constraint for Multirotor eVTOL in Urban Air Mobility,”Journal of Aerospace Information Systems, Vol. 16, No. 7, 2019, pp. 263–277. https://doi.org/10.2514/1.I010710. 45

    2019 doi

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.