Pith. sign in

REVIEW 3 major objections 3 minor 105 references

Learned graph risk can allocate EV-charging safety authority at scale, with physics choosing the correction.

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 · deepseek-v4-flash

2026-08-04 00:43 UTC pith:USAN2T3M

load-bearing objection Abstract-only read: a promising design for separating safety intervention magnitude from direction, but the fixed-authority baseline is under-documented and the comparative claim needs a closer look. the 3 major comments →

arxiv 2608.00679 v1 pith:USAN2T3M submitted 2026-08-01 cs.AI cs.MA

HetGPS: Scalable Graph Multi-Agent Reinforcement Learning with Physics-Anchored Adaptive Safety for EV Charging

Xiangwei Wang , Nanduni Nimalsiri , Yu Xia , Peng Wang , Saman Halgamuge This is my paper
classification cs.AI cs.MA
keywords multi-agent reinforcement learninggraph neural networkselectric vehicle chargingsafety filterphysics-anchored correctionvoltage regulationzero-shot transferadaptive authority
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 proposes a safety-filtering framework for controlling large populations of network-coupled agents, demonstrated on electric vehicle charging. The claim is that the filter can learn how strongly to intervene (authority) from a graph representation of risk, while a physics model decides which direction corrects the violation. Across networks with 200 to 3,218 EVs, the adaptive filter reduces bus–step voltage violations from up to 7.74% to as low as 0.52% while keeping nearly all departures on time. The same trained policy and risk model, with a fixed 383,702 parameters, transfers zero-shot to larger networks with 16 and 32 transformers. A sympathetic reader would care because this points to a way to make safe RL control tractable for large, topology-varying infrastructure systems.

Core claim

The central discovery is the explicit separation of intervention magnitude from corrective direction in a safety filter. A learned, action-conditioned graph residual model schedules state-dependent intervention authority, and a deterministic physics model supplies the correction direction. Coupled with a parameter-shared heterogeneous graph soft actor-critic policy, this lets a single small model control fleets an order of magnitude larger than a centralized actor, with adaptive safety that outperforms fixed-authority projection on reward and safety scores. The empirical result is that learned graph risk can allocate how much to intervene, while feeder physics anchors which way to correct.

What carries the argument

The key machinery is an adaptive authority filter: an action-conditioned graph residual model that produces an intervention magnitude per agent at each step, separated from a physics-based projection that determines the corrective direction. The graph residual is trained jointly with a parameter-shared heterogeneous graph soft actor-critic policy, and model size stays constant as fleet size grows because parameters are shared across the graph. This separation is what lets the system adjust how aggressively to override the policy without losing the physics correctness of the correction.

Load-bearing premise

The framework's transfer claims hinge on the assumption that the graph encoder and learned authority allocation generalize across network topologies and load regimes, so that a policy trained on one grid size remains safe when the number of transformers, EVs, and constraints grow.

What would settle it

Run the trained 8-transformer policy on a 64-transformer network with a load profile that creates simultaneous peaks at distant substations, and compare voltage violations against the unfiltered policy; if the adaptive filter yields violation rates close to the unfiltered baseline (or worse), the zero-shot transfer claim is falsified.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Voltage violations can be cut by more than half on networks with thousands of EVs while maintaining at least 99% departure success.
  • A single trained model, fixed at 383,702 parameters, can be deployed on networks with 200 to 3,200+ EVs without retraining.
  • Learning the magnitude of intervention is more effective than fixing it, as shown by improved mean reward on all five test networks and lower safety score on four.
  • Zero-shot transfer from eight to sixteen and thirty-two transformer networks keeps violation rates below 1%.

Where Pith is reading between the lines

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

  • The magnitude-vs-direction separation may generalize beyond EV charging to other shared-constraint control problems, such as building HVAC or water distribution, where a physical law can supply the direction and learning supplies only the degree of override.
  • The learned authority map could serve as an interpretability signal: it shows where and when the safe coordinator distrusts the learned policy, highlighting systemic weak spots in the underlying RL policy.
  • Because the model size is independent of fleet size, this architecture could enable on-device or edge-deployed safety filters that scale to millions of agents, provided the graph neighbourhood remains bounded.
  • A natural testable extension is to replace the physics direction module with a learned but physics-constrained direction, isolating how much of the gain comes from the authority learning versus the physical anchor.

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 / 3 minor

Summary. The paper proposes HetGPS, a hybrid MARL framework for EV charging in distribution networks. It couples a parameter-shared heterogeneous graph soft actor-critic policy with a learned graph-based risk model that schedules intervention authority, while a physics model determines the corrective direction. The abstract reports: across five nested distribution networks with 200-3,218 EVs and 100 evaluation days, Adaptive Authority reduces bus-step voltage violations from 3.93-7.74% (unfiltered) to 0.52-3.44%, while maintaining 99.06-100% departure success; relative to a fixed-authority physics-directed projection, it improves mean reward on all five networks and lowers mean safety score on four; a policy trained on the eight-transformer system transfers zero-shot to 16- and 32-transformer systems with 0.57-0.75% violation rates and at least 99.99% departure success. The deployed policy-and-risk model has 383,702 parameters at every scale, while a matched centralized SAC actor is about 170x larger at 3,218 EVs.

Significance. If the empirical claims hold, HetGPS would be a meaningful contribution to safety-filter design for networked multi-agent systems. The separation of intervention magnitude from corrective direction is principled, and the use of a learned graph risk model to allocate authority at scale is novel and practically relevant. The parameter-sharing design that keeps the trainable model size independent of fleet size is a valuable scaling property, and the comparison against a centralized SAC actor shows a large efficiency advantage. The evaluation across five nested networks with 100 days is more extensive than is typical in multi-agent RL safety papers, and the zero-shot transfer test is a useful generalization probe. However, because the full text is not available, the abstract alone does not provide enough methodology to assess whether these claims are technically sound; the comparative baseline, evaluation protocol, uncertainty quantification, and transfer conditions are all unspecified.

major comments (3)
  1. [Abstract — fixed-authority baseline] The central comparative claim — "Relative to the same physics-directed projection with fixed authority" — depends critically on how the fixed authority value was set. The abstract nowhere states whether this baseline was tuned per network, set to a single constant, or chosen arbitrarily. If the fixed authority is an untuned or poorly chosen constant, the adaptive method's advantage could be trivially explained by occasionally deviating from a bad baseline, rather than by learned graph risk capturing meaningful safety structure. The full paper must specify the baseline selection procedure, the value(s) used, and ideally a sensitivity analysis over the fixed authority setting.
  2. [Abstract — evaluation protocol and uncertainty] All reported safety and success numbers are point estimates over 100 evaluation days with no standard deviations, confidence intervals, or significance tests. It is also unclear what exactly a "bus-step voltage violation" counts (per bus per time step? per event?) and how "departure success" is defined. Without a precise definition of the metrics and the simulation protocol (time step, load and EV arrival processes, stochastic seeds), the reported ranges 0.52-3.44% and 99.06-100% cannot be reproduced or interpreted. The full paper should provide a complete evaluation protocol and uncertainty quantification.
  3. [Abstract — zero-shot transfer] The zero-shot transfer claim — training on the eight-transformer system and transferring to 16- and 32-transformer systems — assumes that the graph encoder and parameter sharing capture topology-invariant structure and that the learned authority allocation remains appropriate under network growth. The abstract does not describe the generation of the test networks, their load and EV profiles, or the degree of distribution shift. Without this information, the impressive transfer numbers (0.57-0.75% violations, >=99.99% departure success) are not interpretable. The full paper must document the network generation procedure and the similarity/dissimilarity between training and test distributions.
minor comments (3)
  1. [Abstract — safety improvement asymmetry] The abstract says the method "improves mean reward on all five networks" but "lowers the mean safety score on four" networks. It would be useful to explain why the safety score does not improve on the fifth network, and whether that network corresponds to a case where the fixed-authority baseline was already near-optimal or where the learned authority can degrade safety.
  2. [Abstract — model size independence] The statement that the deployed policy-and-risk model contains 383,702 learned parameters "at every scale" is potentially misleading if input/output encoders are per-fleet or per-agent. Please clarify whether this count excludes any network-size-dependent embedding layers and what exactly constitutes the "model."
  3. [Abstract — centralized SAC comparison] The comparison with a "matched centralized SAC actor" should state whether the same architecture, hyperparameters, and training budget were used for the centralized baseline, and why the parameter-count comparison (170x) is the relevant metric rather than task performance or compute time.

Circularity Check

0 steps flagged

No circularity found; the comparison and zero-shot transfer are independent of the fitted components by construction.

full rationale

This is an abstract-only review, and the abstract describes an empirical comparison rather than a derivation that could reduce to its inputs. The central claim is that a learned graph residual model schedules intervention authority while a physics model determines corrective direction, and this hybrid filter is compared against (i) no filtering and (ii) the same physics-directed projection with fixed authority. Neither comparison defines the learned authority in terms of the outcome metric, nor is the fixed-authority baseline constructed from the learned schedule. The zero-shot transfer from an 8-transformer system to 16- and 32-transformer systems is a genuine extrapolation to unseen network sizes, not a prediction of fitted values on the same data. No self-citation is invoked as load-bearing, no uniqueness theorem is imported, and no ansatz is smuggled in via citation. The skeptical concern that the fixed-authority baseline may be untuned is a legitimate experimental-design question about the fairness of the comparison, but it is not circularity: even an unfair baseline would not make the learned component definitionally equal to the reported improvement. Without equations or a derivation chain, there is no exhibited reduction of a claimed result to its own inputs. The honest finding is no significant circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The abstract introduces no new physical entities or free parameters by name; the learned parameters of the policy/risk model (383,702) are trained, not fitted to a single target. However, two domain assumptions are load-bearing: the accuracy of the feeder physics model and the topology invariance of the graph representation. Without full methodology, the ledger cannot be exhaustively completed.

axioms (2)
  • domain assumption The feeder physics model accurately computes voltage constraints and corrective directions.
    The filter's corrective direction is determined by a physics model; if that model is wrong or too coarse, the safety guarantees would erode. This is stated implicitly in the abstract's description of 'physics-anchored correction'.
  • domain assumption The graph representation and parameter sharing are invariant enough across transformer scales for zero-shot transfer.
    The transfer claim from the eight-transformer system to 16- and 32-transformer systems assumes that the learned graph policy captures scale-invariant topology and load patterns. The abstract gives no evidence for this invariance beyond the reported transfer numbers.

pith-pipeline@v1.3.0-daily-deepseek · 598 in / 5424 out tokens · 52134 ms · 2026-08-04T00:43:09.415810+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of HetGPS: Scalable Graph Multi-Agent Reinforcement Learning with Physics-Anchored Adaptive Safety for EV Charging." pith.science (2026). https://pith.science/paper/USAN2T3M

@misc{pith2026260800679,
  author       = {Pith},
  title        = {Pith review of: HetGPS: Scalable Graph Multi-Agent Reinforcement Learning with Physics-Anchored Adaptive Safety for EV Charging},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/USAN2T3M}},
  note         = {Machine review of arXiv:2608.00679}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Safety interventions for large populations of network-coupled agents must protect shared constraints without unnecessarily overriding task-oriented policy decisions. We present HetGPS, a hybrid graph-control framework synergizing learned graph risk with physics-anchored correction by separating intervention magnitude from corrective direction. An action-conditioned graph residual model schedules state-dependent intervention authority, while a physics model determines its direction. For electric vehicle (EV) charging, we couple this filter with a parameter-shared heterogeneous graph soft actor-critic policy, enabling topology-aware coordination with a learned model size independent of fleet size. Across five nested distribution networks with 200--3,218 EVs and 100 evaluation days, Adaptive Authority reduces bus--step voltage violations from 3.93--7.74\% without filtering to 0.52--3.44\%, while maintaining 99.06--100\% departure success. Relative to the same physics-directed projection with fixed authority, it improves mean reward on all five networks and lowers the mean safety score on four. The deployed policy-and-risk model contains 383,702 learned parameters at every scale; at 3,218 EVs, a matched centralized SAC actor is about $170\times$ larger. A policy trained on the eight-transformer system transfers zero-shot to the 16- and 32-transformer systems, attaining 0.57--0.75\% violation rates and at least 99.99\% departure success. These results show that learned graph risk can allocate intervention authority at scale while feeder physics anchors corrective action.

Figures

Figures reproduced from arXiv: 2608.00679 by Nanduni Nimalsiri, Peng Wang, Saman Halgamuge, Xiangwei Wang, Yu Xia.

Figure 1
Figure 1. Figure 1: End-to-end EV-control pipeline. Offline, the shared [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Policy-training reward per agent (EV) over 1,000 [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Spatial and temporal intervention on a represen [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 3
Figure 3. Figure 3: Reward–safety comparison using the three-run [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Target-normalized topology transfer from the K8 controller. Panel (a) uses [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Full-network visualization of the heterogeneous [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Physics-anchor validation on K8 and K32. Direc [PITH_FULL_IMAGE:figures/full_fig_p017_7.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

105 extracted references

  1. [1]

    Thurner and A

    L. Thurner and A. Scheidler and F. Schafer and J.-H. Menke and J. Dollichon and F. Meier and S. Meinecke and M. Braun , title =. IEEE Trans. Power Syst. , volume =. 2018 , doi =

    2018

  2. [2]

    M. E. Baran and F. F. Wu , title =. IEEE Trans. Power Deliv. , volume =. 1989 , doi =

    1989

  3. [3]

    N. I. Nimalsiri and E. L. Ratnam and C. P. Mediwaththe and D. B. Smith and S. K. Halgamuge , title =. Appl. Energy , volume =. 2021 , doi =

    2021

  4. [4]

    Haarnoja and A

    T. Haarnoja and A. Zhou and P. Abbeel and S. Levine , title =. Proc. Int. Conf. Mach. Learn. (ICML) , pages =

  5. [5]

    Chis and J

    A. Chis and J. Lunden and V. Koivunen , title =. IEEE Trans. Veh. Technol. , volume =. 2017 , doi =

    2017

  6. [6]

    Wan and H

    Z. Wan and H. Li and H. He and D. Prokhorov , title =. IEEE Trans. Smart Grid , volume =. 2019 , doi =

    2019

  7. [7]

    Wang and S

    S. Wang and S. Bi and Y.-J. A. Zhang , title =. Proc. IEEE Power Energy Soc. Gen. Meeting (PESGM) , pages =. 2018 , doi =

    2018

  8. [8]

    Wang and S

    S. Wang and S. Bi and Y.-J. A. Zhang , title =. IEEE Trans. Ind. Informat. , volume =. 2021 , doi =

    2021

  9. [9]

    Yan and X

    L. Yan and X. Chen and Y. Chen and J. Wen , title =. IEEE Trans. Ind. Informat. , volume =. 2022 , doi =

    2022

  10. [10]

    F. L. Da Silva and C. E. H. Nishida and D. M. Roijers and A. H. R. Costa , title =. IEEE Trans. Smart Grid , volume =. 2020 , doi =

    2020

  11. [11]

    Lowe and Y

    R. Lowe and Y. Wu and A. Tamar and J. Harb and P. Abbeel and I. Mordatch , title =. Proc. Adv. Neural Inf. Process. Syst. (NeurIPS) , pages =

  12. [12]

    Li and Z

    H. Li and Z. Wan and H. He , title =. IEEE Trans. Smart Grid , volume =. 2020 , doi =

    2020

  13. [13]

    Liang and F

    Q. Liang and F. Que and E. Modiano , title =

  14. [14]

    Stooke and J

    A. Stooke and J. Achiam and P. Abbeel , title =. Proc. Int. Conf. Mach. Learn. (ICML) , pages =

  15. [15]

    K. P. Wabersich and M. N. Zeilinger , title =. Automatica , volume =. 2021 , doi =

    2021

  16. [16]

    Paesschesoone and N

    S. Paesschesoone and N. Kayedpour and C. Manna and G. Crevecoeur , title =. 2024 , doi =

    2024

  17. [17]

    T. N. Kipf and M. Welling , title =. Proc. Int. Conf. Learn. Represent. (ICLR) , year =

  18. [18]

    Hamilton and Z

    W. Hamilton and Z. Ying and J. Leskovec , title =. Proc. Adv. Neural Inf. Process. Syst. (NeurIPS) , pages =

  19. [19]

    Owerko and F

    D. Owerko and F. Gama and A. Ribeiro , title =. Proc. IEEE Int. Conf. Acoust., Speech Signal Process. (ICASSP) , pages =. 2020 , doi =

    2020

  20. [20]

    R. R. Hossain and Q. Huang and R. Huang , title =. IEEE Trans. Power Syst. , volume =. 2021 , doi =

    2021

  21. [21]

    Zhang and M

    Y. Zhang and M. Yue and J. Wang and S. Yoo , title =. IEEE Trans. Neural Netw. Learn. Syst. , volume =. 2024 , doi =

    2024

  22. [22]

    Cao and J

    D. Cao and J. Zhao and J. Hu and Y. Pei and Q. Huang and Z. Chen and W. Hu , title =. IEEE Trans. Smart Grid , volume =. 2024 , doi =

    2024

  23. [23]

    Yan and Q

    R. Yan and Q. Xing and Y. Xu , title =. IEEE Trans. Smart Grid , volume =. 2024 , doi =

    2024

  24. [24]

    Pan and C

    W. Pan and C. Liu , title =. Int. J. Comput. Commun. Control , volume =. 2023 , doi =

    2023

  25. [25]

    Rampasek and M

    L. Rampasek and M. Galkin and V. P. Dwivedi and A. T. Luu and G. Wolf and D. Beaini , title =. Proc. Adv. Neural Inf. Process. Syst. (NeurIPS) , volume =

  26. [26]

    Zhao and B

    Y. Zhao and B. Zhang , title =. Advanced Data Analytics for Power Systems , editor =. 2021 , doi =

    2021

  27. [27]

    E. L. Ratnam and S. R. Weller and C. M. Kellett and A. T. Murray , title =. Int. J. Sustain. Energy , volume =. 2017 , doi =

    2017

  28. [28]

    2023 , note =

    Aggregated price and demand data--. 2023 , note =

    2023

  29. [29]

    Powell and G

    S. Powell and G. V. Cezar and L. Min and I. M. L. Azevedo and R. Rajagopal , title =. Nature Energy , volume =. 2022 , doi =

    2022

  30. [30]

    Qiu and Y

    D. Qiu and Y. Wang and W. Hua and G. Strbac , title =. Renew. Sustain. Energy Rev. , volume =. 2023 , doi =

    2023

  31. [31]

    M. A. Ortega-Vazquez , title =. IET Gener., Transmiss. Distrib. , volume =. 2014 , doi =

    2014

  32. [32]

    Farzin and M

    H. Farzin and M. Fotuhi-Firuzabad and M. Moeini-Aghtaie , title =. IEEE Trans. Sustain. Energy , volume =. 2016 , doi =

    2016

  33. [33]

    Shafie-Khah and P

    M. Shafie-Khah and P. Siano and D. Z. Fitiwi and N. Mahmoudi and J. P. S. Catal. IEEE Transactions on Smart Grid , volume =. 2018 , doi =

    2018

  34. [34]

    Alfaverh and M

    F. Alfaverh and M. Denai and Y. Sun , title =. Electr. Power Syst. Res. , volume =. 2023 , doi =

    2023

  35. [35]

    Celik and R

    B. Celik and R. Roche and S. Suryanarayanan and D. Bouquain and A. Miraoui , title =. Renew. Sustain. Energy Rev. , volume =. 2017 , doi =

    2017

  36. [36]

    Kuang and H

    H. Kuang and H. Qu and K. Deng and J. Li , title =. Appl. Energy , volume =. 2024 , doi =

    2024

  37. [37]

    Zhao and C

    Z. Zhao and C. K. M. Lee and J. Ren , title =. Appl. Energy , volume =. 2024 , doi =

    2024

  38. [38]

    Nimalsiri and E

    N. Nimalsiri and E. Ratnam and D. Smith and C. Mediwaththe and S. Halgamuge , title =. IEEE Trans. Ind. Informat. , volume =. 2024 , doi =

    2024

  39. [39]

    Liu and P

    M. Liu and P. K. Phanivong and Y. Shi and D. S. Callaway , title =. IEEE Trans. Control Syst. Technol. , volume =. 2019 , doi =

    2019

  40. [40]

    Ma and D

    Z. Ma and D. S. Callaway and I. A. Hiskens , title =. IEEE Trans. Control Syst. Technol. , volume =. 2013 , doi =

    2013

  41. [41]

    Duan and D

    J. Duan and D. Shi and R. Diao and H. Li and Z. Wang and B. Zhang and D. Bian and Z. Yi , title =. IEEE Trans. Power Syst. , volume =. 2020 , doi =

    2020

  42. [42]

    Ernst and M

    D. Ernst and M. Glavic and F. Capitanescu and L. Wehenkel , title =. IEEE Trans. Syst., Man, Cybern. B, Cybern. , volume =. 2009 , doi =

    2009

  43. [43]

    Qin and D

    Z. Qin and D. Liu and H. Hua and J. Cao , title =. IEEE Trans. Smart Grid , volume =. 2021 , doi =

    2021

  44. [44]

    Yaniv and P

    A. Yaniv and P. Kumar and Y. Beck , title =. Electr. Power Syst. Res. , volume =. 2023 , doi =

    2023

  45. [45]

    Lin and Y

    H. Lin and Y. Sun , title =. Proc. IEEE Global Commun. Conf. (GLOBECOM) , pages =. 2022 , doi =

    2022

  46. [46]

    Gronauer and K

    S. Gronauer and K. Diepold , title =. Artif. Intell. Rev. , volume =. 2022 , doi =

    2022

  47. [47]

    Huang and Y

    R. Huang and Y. Chen and T. Yin and X. Li and A. Li and J. Tan and W. Yu and Y. Liu and Q. Huang , title =. IEEE Trans. Power Syst. , volume =. 2022 , doi =

    2022

  48. [48]

    Velickovic and G

    P. Velickovic and G. Cucurull and A. Casanova and A. Romero and P. Lio and Y. Bengio , title =. Proc. Int. Conf. Learn. Represent. (ICLR) , year =

  49. [49]

    Liao and B

    W. Liao and B. Bak-Jensen and J. R. Pillai and Y. Wang and Y. Wang , title =. J. Mod. Power Syst. Clean Energy , volume =. 2022 , doi =

    2022

  50. [50]

    Haarnoja and A

    T. Haarnoja and A. Zhou and K. Hartikainen and G. Tucker and S. Ha and J. Tan and V. Kumar and H. Zhu and A. Gupta and P. Abbeel and S. Levine , title =

  51. [51]

    A. N. Angelopoulos and S. Bates , title =. Foundations and Trends in Machine Learning , volume =. 2023 , doi =

    2023

  52. [52]

    Romano and E

    Y. Romano and E. Patterson and E. J. Cand\`es , title =. Advances in Neural Information Processing Systems , volume =

  53. [53]

    Achiam and D

    J. Achiam and D. Held and A. Tamar and P. Abbeel , title =. Proc. Int. Conf. Mach. Learn. (ICML) , volume =

  54. [54]

    Lu and K

    S. Lu and K. Zhang and T. Chen and T. Ba. Decentralized Policy Gradient Descent Ascent for Safe Multi-Agent Reinforcement Learning , booktitle =. 2021 , doi =

    2021

  55. [55]

    Alshiekh and R

    M. Alshiekh and R. Bloem and R. Ehlers and B. K. Safe Reinforcement Learning via Shielding , booktitle =. 2018 , doi =

    2018

  56. [56]

    Yu and W

    H. Yu and W. Xu and H. Zhang , title =. Proc. Adv. Neural Inf. Process. Syst. (NeurIPS) , volume =. 2022 , doi =

    2022

  57. [57]

    Tian and H

    H. Tian and H. Hamedmoghadam and R. Shorten and P. Ferraro , title =. Proc. Adv. Neural Inf. Process. Syst. (NeurIPS) , volume =. 2024 , doi =

    2024

  58. [58]

    Jiang and C

    J. Jiang and C. Dun and T. Huang and Z. Lu , title =. Proc. Int. Conf. Learn. Represent. (ICLR) , year =

  59. [59]

    Wang and R

    T. Wang and R. Liao and J. Ba and S. Fidler , title =. Proc. Int. Conf. Learn. Represent. (ICLR) , year =

  60. [60]

    Fu and T

    Q. Fu and T. Qiu and J. Yi and Z. Pu and S. Wu , title =. Proc. AAAI Conf. Artif. Intell. (AAAI) , volume =. 2022 , doi =

    2022

  61. [61]

    Fan and Z

    W. Fan and Z. Yu and C. Ma and C. Li and Y. Yang and X. Zhang , title =. Proc. AAAI Conf. Artif. Intell. (AAAI) , volume =. 2025 , doi =

    2025

  62. [62]

    Wang and W

    J. Wang and W. Xu and Y. Gu and W. Song and T. C. Green , title =. Proc. Adv. Neural Inf. Process. Syst. (NeurIPS) , volume =

  63. [63]

    Qu and J

    Y. Qu and J. Ma and F. Wu , title =. Proc. Int. Joint Conf. Artif. Intell. (IJCAI) , pages =. 2024 , doi =

    2024

  64. [64]

    Yeh and V

    C. Yeh and V. Li and R. Datta and J. Arroyo and N. Christianson and C. Zhang and Y. Chen and M. M. Hosseini and A. Golmohammadi and Y. Shi and Y. Yue and A. Wierman , title =. Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), Datasets and Benchmarks Track , volume =. 2023 , doi =

    2023

  65. [65]

    Liu and R

    F. Liu and R. Sen and J. P. Talusan and A. Pettet and A. Kandel and Y. Suzue and A. Mukhopadhyay and A. Dubey , title =. Proc. Int. Conf. Auton. Agents Multiagent Syst. (AAMAS) , pages =

  66. [66]

    Valogianni and W

    K. Valogianni and W. Ketter and J. Collins and D. Zhdanov , title =. Proc. AAAI Conf. Artif. Intell. (AAAI) , volume =. 2014 , doi =

    2014

  67. [67]

    Achiam, J.; Held, D.; Tamar, A.; and Abbeel, P. 2017. Constrained Policy Optimization. In Proc. Int. Conf. Mach. Learn. (ICML), volume 70, 22--31

    2017

  68. [68]

    Alshiekh, M.; Bloem, R.; Ehlers, R.; K \"o nighofer, B.; Niekum, S.; and Topcu, U. 2018. Safe Reinforcement Learning via Shielding. In Proc. AAAI Conf. Artif. Intell. (AAAI), volume 32, 2669--2678

    2018

  69. [69]

    Australian Energy Market Operator . 2023. Aggregated price and demand data-- Victoria ( VIC1 ). [Online]. Available: https://aemo.com.au/energy-systems/electricity/national-electricity-market-nem/data-nem/aggregated-data. Accessed: Jul. 3, 2026

    2023

  70. [70]

    E.; and Wu, F

    Baran, M. E.; and Wu, F. F. 1989. Network reconfiguration in distribution systems for loss reduction and load balancing. IEEE Trans. Power Deliv., 4(2): 1401--1407

    1989

  71. [71]

    Cao, D.; Zhao, J.; Hu, J.; Pei, Y.; Huang, Q.; Chen, Z.; and Hu, W. 2024. Physics-informed graphical representation-enabled deep reinforcement learning for robust distribution system voltage control. IEEE Trans. Smart Grid, 15(1): 233--246

    2024

  72. [72]

    Department of Transport and Planning, Victoria . 2020. Victorian Integrated Survey of Travel and Activity ( VISTA ). [Online]. Available: https://transport.vic.gov.au/about/data-and-research/vista. Accessed: Jul. 3, 2026

    2020

  73. [73]

    Fan, W.; Yu, Z.; Ma, C.; Li, C.; Yang, Y.; and Zhang, X. 2025. Towards Efficient Collaboration via Graph Modeling in Reinforcement Learning. In Proc. AAAI Conf. Artif. Intell. (AAAI), volume 39, 16505--16513

    2025

  74. [74]

    Farzin, H.; Fotuhi-Firuzabad, M.; and Moeini-Aghtaie, M. 2016. A Practical Scheme to Involve Degradation Cost of Lithium-Ion Batteries in Vehicle-to-Grid Applications. IEEE Trans. Sustain. Energy, 7(4): 1730--1738

    2016

  75. [75]

    Fu, Q.; Qiu, T.; Yi, J.; Pu, Z.; and Wu, S. 2022. Concentration Network for Reinforcement Learning of Large-Scale Multi-Agent Systems. In Proc. AAAI Conf. Artif. Intell. (AAAI), volume 36, 9341--9349

    2022

  76. [76]

    Haarnoja, T.; Zhou, A.; Abbeel, P.; and Levine, S. 2018. Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor. In Proc. Int. Conf. Mach. Learn. (ICML), 1861--1870

    2018

  77. [77]

    Jiang, J.; Dun, C.; Huang, T.; and Lu, Z. 2020. Graph Convolutional Reinforcement Learning. In Proc. Int. Conf. Learn. Represent. (ICLR)

    2020

  78. [78]

    Li, H.; Wan, Z.; and He, H. 2020. Constrained EV charging scheduling based on safe deep reinforcement learning. IEEE Trans. Smart Grid, 11(3): 2427--2439

    2020

  79. [79]

    P.; Pettet, A.; Kandel, A.; Suzue, Y.; Mukhopadhyay, A.; and Dubey, A

    Liu, F.; Sen, R.; Talusan, J. P.; Pettet, A.; Kandel, A.; Suzue, Y.; Mukhopadhyay, A.; and Dubey, A. 2025. Reinforcement Learning-Based Approach for Vehicle-to-Building Charging with Heterogeneous Agents and Long-Term Rewards. In Proc. Int. Conf. Auton. Agents Multiagent Syst. (AAMAS), 1345--1353

    2025

  80. [80]

    K.; Shi, Y.; and Callaway, D

    Liu, M.; Phanivong, P. K.; Shi, Y.; and Callaway, D. S. 2019. Decentralized charging control of electric vehicles in residential distribution networks. IEEE Trans. Control Syst. Technol., 27(1): 266--281

    2019

Showing first 80 references.