pith. sign in

arxiv: 2605.16547 · v1 · pith:XZE66VDDnew · submitted 2026-05-15 · 💻 cs.LG

World Model-Enabled Causal Digital Twins for Semantic Communications in Physical AI Systems

Lingyi Wang , Tingyu Shui , Walid Saad , Pascal Adjakple This is my paper

Pith reviewed 2026-05-20 19:37 UTC · model grok-4.3

classification 💻 cs.LG
keywords semantic communicationscausal digital twinsworld modelsphysical AI systemsclosed-loop controlUAV navigationreinforcement learningcounterfactual reasoning
0
0 comments X

The pith

A world-model causal digital twin framework improves long-term return per bit in semantic communications for closed-loop physical AI systems.

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

This paper tackles semantic communications in physical AI systems where transmitted information shapes not just immediate inferences but ongoing control actions and state evolution over long horizons. It reformulates the problem as maximizing expected return per transmitted bit under wireless constraints and introduces a causal information value metric to rank each semantic token by its contribution to future returns. The WM-CDT framework then deploys a learned world model to generate counterfactual imagined rollouts, training an actor-critic policy and token selector with far less real interaction data. A sympathetic reader would care because conventional semantic methods optimize one-shot tasks and break down when communication and control must remain coupled across time.

Core claim

The authors formulate semantic communications as long-term return-per-bit maximization in closed-loop sensing-communication-inference-control systems and solve it with a world-model-enabled causal digital twin that performs counterfactual reasoning on imagined trajectories, yielding an actor-critic control policy and a CIV-per-bit semantic token selector that together raise return-per-kbit and navigation success rate over standard reinforcement learning baselines in AirSim-Sionna UAV simulations.

What carries the argument

The causal information value (CIV) metric, which measures the marginal effect of transmitting a given semantic token on expected long-term return through hypothetical transmission interventions, paired with world-model-generated imagined rollouts for high-data-efficiency policy training.

If this is right

  • Semantic token selection can be performed by ranking tokens according to their CIV per transmitted bit while respecting wireless budgets.
  • Control policies can be trained from imagined trajectories rather than costly real-world interactions, raising sample efficiency.
  • Joint optimization of communication and control yields higher long-horizon task success than myopic, one-shot semantic approaches.
  • The framework directly supports goal-oriented networking in any closed-loop physical AI setting with bit-rate limits.

Where Pith is reading between the lines

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

  • The same counterfactual rollout technique could be applied to other bandwidth-limited physical AI tasks such as autonomous driving or multi-robot coordination.
  • If the world model drifts from reality, periodic online fine-tuning of the digital twin may be needed to preserve performance.
  • This approach suggests that future semantic communication standards should include explicit support for causal intervention metrics rather than relying solely on reconstruction fidelity.

Load-bearing premise

The learned world model inside the causal digital twin must accurately reproduce the true dynamics of the closed-loop physical AI system so that its counterfactual rollouts remain reliable guides for policy improvement.

What would settle it

Deploy the WM-CDT policy in the same AirSim-Sionna UAV navigation simulator and measure whether return-per-kbit or navigation success rate fails to exceed the performance of existing reinforcement learning baselines.

Figures

Figures reproduced from arXiv: 2605.16547 by Lingyi Wang, Pascal Adjakple, Tingyu Shui, Walid Saad.

Figure 1
Figure 1. Figure 1: The closed-loop goal-oriented semantic communication system with a world model-enabled causal digital twin. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The proposed AirSim-Sionna-based simulator for semantic commu [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The training losses of different components in WM-CDT. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The convergence of different methods with limited environment steps. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: The navigation success rate of different approaches under bit budgets. [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 5
Figure 5. Figure 5: The return-per-kbit of different approaches under bit budgets. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 9
Figure 9. Figure 9: Ablation study of the latent dynamics design under different imagi [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Ablation study of the semantic encoder design under different [PITH_FULL_IMAGE:figures/full_fig_p011_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Ablation study of token-selection strategies under different imagi [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
read the original abstract

Semantic communication has emerged as a promising paradigm for enabling goal-oriented networking. However, most existing semantic communication solutions are tailored to one-shot tasks and optimize instantaneous performance. Hence, they cannot be used to support closed-loop dynamic systems with physical artificial intelligence (AI), in which the transmitted semantics affect not only the current inference outcome but also future control actions, state evolution, and ultimately long-horizon task performance. To address this gap, this paper investigates goal-oriented semantic communications for physical AI systems with closed-loop sensing-communication-inference-control. In particular, the problem of semantic communications is formulated as a long-term return-per-bit maximization under wireless bit-budget constraints while capturing both control efficiency and communication efficiency. To solve this problem, a novel causal information value (CIV) metric is introduced to evaluate the marginal contribution of each semantic token to the expected long-term return by transmission interventions. Then, a world-model-enabled causal digital twin (WM-CDT) framework is proposed to capture the dynamics of closed-loop physical AI systems and enable counterfactual reasoning for long-horizon imagined rollouts. Based on these imagined rollouts, an actor-critic policy is trained for long-horizon agent control with high data efficiency, while the semantic token selector is trained through CIV-per-bit evaluation. Extensive simulations on an AirSim-Sionna-based unmanned aerial vehicle (UAV) navigation simulator show that the proposed WM-CDT framework achieves significant improvement in return-per-kbit and navigation success rate compared to existing reinforcement learning solutions.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper formulates goal-oriented semantic communications for closed-loop physical AI systems as a long-term return-per-bit maximization problem under wireless bit-budget constraints. It introduces a Causal Information Value (CIV) metric that quantifies each semantic token's marginal contribution to expected long-term return via transmission interventions, and proposes a World Model-Enabled Causal Digital Twin (WM-CDT) that learns system dynamics to support counterfactual reasoning and imagined rollouts. An actor-critic policy is trained on these rollouts for high-data-efficiency control while the token selector uses CIV-per-bit; AirSim-Sionna UAV navigation simulations are reported to show gains in return-per-kbit and navigation success rate over standard RL baselines.

Significance. If the world-model fidelity and counterfactual validity hold, the framework could meaningfully advance semantic communications for dynamic physical systems by jointly optimizing communication and long-horizon control efficiency. The combination of causal interventions, imagined rollouts, and actor-critic training on a digital twin is a coherent direction that addresses limitations of one-shot semantic comm approaches. The simulation results, if substantiated, would constitute a concrete demonstration of improved data efficiency in a realistic UAV setting.

major comments (2)
  1. [§4] §4 (AirSim-Sionna experiments): the headline claims of improved return-per-kbit and navigation success rest on the WM-CDT producing reliable imagined trajectories under semantic-token interventions, yet no multi-step prediction error (e.g., position/velocity MSE over 10–20 steps) or sensitivity analysis to model mismatch is reported. If learned dynamics deviate from true simulator transitions outside the training distribution, the counterfactual returns used for policy optimization become biased, directly undermining the superiority over standard RL baselines.
  2. [Abstract, §3] Abstract and §3 (CIV definition): CIV is defined as the marginal contribution of each token to expected long-term return, which is then optimized inside the same actor-critic loop that fits the policy; this creates a dependency that may render the metric non-independent of the fitted policy and requires explicit justification or ablation to confirm it does not inflate reported gains.
minor comments (2)
  1. [§4] The manuscript should specify the exact number of simulator runs, random seeds, and data-exclusion rules used to generate the performance figures; error bars or confidence intervals are also missing from the reported metrics.
  2. [§3] Notation for the world-model transition function and the intervention operator in the CIV definition should be made fully explicit (e.g., distinguishing p(s'|s,a) from the learned ˆp) to allow readers to reproduce the counterfactual rollout procedure.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major comment below, indicating planned revisions to strengthen the manuscript's rigor and clarity.

read point-by-point responses

  1. Referee: [§4] §4 (AirSim-Sionna experiments): the headline claims of improved return-per-kbit and navigation success rest on the WM-CDT producing reliable imagined trajectories under semantic-token interventions, yet no multi-step prediction error (e.g., position/velocity MSE over 10–20 steps) or sensitivity analysis to model mismatch is reported. If learned dynamics deviate from true simulator transitions outside the training distribution, the counterfactual returns used for policy optimization become biased, directly undermining the superiority over standard RL baselines.

    Authors: We agree that explicit validation of multi-step predictive accuracy and robustness to model mismatch would strengthen the claims regarding the reliability of imagined rollouts. In the revised version, we will add a dedicated subsection in §4 reporting position and velocity MSE for 10–20 step predictions on held-out trajectories from the AirSim-Sionna simulator. We will also include a sensitivity analysis that perturbs world-model parameters (e.g., by training on subsets of data or adding noise) and evaluates the resulting impact on navigation success rate and return-per-kbit. These additions will directly address potential bias in counterfactual returns. revision: yes

  2. Referee: [Abstract, §3] Abstract and §3 (CIV definition): CIV is defined as the marginal contribution of each token to expected long-term return, which is then optimized inside the same actor-critic loop that fits the policy; this creates a dependency that may render the metric non-independent of the fitted policy and requires explicit justification or ablation to confirm it does not inflate reported gains.

    Authors: The CIV metric is computed via counterfactual interventions on the world-model dynamics, which are learned to approximate system transitions independently of the specific policy parameters. Nevertheless, we acknowledge the need for explicit justification of independence in the joint optimization setting. We will revise §3 to clarify the separation between world-model training and policy optimization, and add an ablation study in §4 that compares performance when CIV is evaluated using a frozen world model versus the jointly updated one. This will confirm that the reported gains are not artifacts of the dependency. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper introduces CIV as a marginal contribution metric computed via interventions on the world-model rollouts, then trains the token selector and actor-critic policy on those same imagined trajectories. No equation or definition reduces the claimed return-per-kbit gains or navigation success improvements to a tautological fit or self-citation by construction. The central results rest on explicit comparisons against RL baselines inside the AirSim-Sionna simulator, which supplies an external benchmark independent of the fitted CIV values. The framework therefore remains self-contained against external evaluation rather than internally forced.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 2 invented entities

Core contributions rest on newly introduced CIV metric and WM-CDT as invented entities, plus domain assumptions about closed-loop dynamics and bit constraints.

free parameters (1)
  • wireless bit-budget constraints
    Used to constrain transmission but specific values or fitting process not detailed in abstract.
axioms (1)
  • domain assumption Transmitted semantics affect not only current inference but also future control actions and state evolution in closed-loop systems.
    Invoked in the problem formulation to justify long-horizon optimization.
invented entities (2)
  • Causal Information Value (CIV) metric no independent evidence
    purpose: Evaluate marginal contribution of each semantic token to expected long-term return via transmission interventions.
    Newly defined in the paper without external benchmarks.
  • World-model-enabled causal digital twin (WM-CDT) no independent evidence
    purpose: Capture dynamics of closed-loop physical AI systems and enable counterfactual reasoning for imagined rollouts.
    Proposed framework without independent validation outside the simulations.

pith-pipeline@v0.9.0 · 5807 in / 1282 out tokens · 59257 ms · 2026-05-20T19:37:38.088771+00:00 · methodology

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

43 extracted references · 43 canonical work pages

  1. [1]

    Artificial general intelligence (AGI)-native wireless systems: A journey beyond 6G,

    W. Saad, O. Hashash, C. K. Thomas, C. Chaccour, M. Debbah, N. Man- dayam, and Z. Han, “Artificial general intelligence (AGI)-native wireless systems: A journey beyond 6G,”Proceedings of the IEEE, vol. 113, no. 9, pp. 849–887, Sept. 2025

    work page 2025

  2. [2]

    Energy-efficient edge inference in integrated sensing, communication, and computation networks,

    J. Yao, W. Xu, G. Zhu, K. Huang, and S. Cui, “Energy-efficient edge inference in integrated sensing, communication, and computation networks,”IEEE J. Sel. Areas Commun., vol. 43, no. 10, pp. 3580–3595, Oct. 2025

    work page 2025

  3. [3]

    Bridging physical and digital worlds: embodied large ai for future wireless systems,

    X. Wang, F. Zhu, Z. Yang, C. Huang, X. Chen, Z. Zhang, S. Muhaidat, and M. Debbah, “Bridging physical and digital worlds: embodied large ai for future wireless systems,”arXiv preprint arXiv:2506.24009, 2025

    work page arXiv 2025

  4. [4]

    Less data, more knowledge: Building next-generation semantic communication networks,

    C. Chaccour, W. Saad, M. Debbah, Z. Han, and H. V . Poor, “Less data, more knowledge: Building next-generation semantic communication networks,”IEEE Commun. Surveys Tuts., vol. 27, no. 1, pp. 37–76, Feb. 2024

    work page 2024

  5. [5]

    Adaptive resource allocation for semantic communication networks,

    L. Wang, W. Wu, F. Zhou, Z. Yang, Z. Qin, and Q. Wu, “Adaptive resource allocation for semantic communication networks,”IEEE Trans. Commun., vol. 72, no. 11, pp. 6900–6916, Nov. 2024

    work page 2024

  6. [6]

    Intellicise wireless networks from semantic communications: A survey, research issues, and challenges,

    P. Zhang, W. Xu, Y . Liu, X. Qin, K. Niu, S. Cui, G. Shi, Z. Qin, X. Xu, F. Wanget al., “Intellicise wireless networks from semantic communications: A survey, research issues, and challenges,”IEEE Commun. Surveys Tuts., vol. 27, no. 3, pp. 2051–2084, Jun. 2024

    work page 2051

  7. [7]

    Sensing-communication-computing-control closed-loop opti- mization for 6G digital twin-empowered robotic systems,

    X. Fang, C. Lei, W. Feng, Y . Chen, M. Xiao, N. Ge, and C.- X. Wang, “Sensing-communication-computing-control closed-loop opti- mization for 6G digital twin-empowered robotic systems,”IEEE J. Sel. Areas Commun., vol. 43, no. 10, pp. 3330–3346, Oct. 2025

    work page 2025

  8. [8]

    Feature importance-aware task-oriented semantic transmission and op- timization,

    Y . Wang, S. Han, X. Xu, H. Liang, R. Meng, C. Dong, and P. Zhang, “Feature importance-aware task-oriented semantic transmission and op- timization,”IEEE Trans. on Cogn. Commun. Netw., vol. 10, no. 4, pp. 1175–1189, Aug. 2024

    work page 2024

  9. [9]

    Age-of- information vs. value-of-information scheduling for cellular networked control systems,

    O. Ayan, M. Vilgelm, M. Kl ¨ugel, S. Hirche, and W. Kellerer, “Age-of- information vs. value-of-information scheduling for cellular networked control systems,” inProc. ACM/IEEE Int. Conf. Cyber-Phys. Syst. (ICCPS), Montreal, Canada, Apr. 2019, pp. 109–117

    work page 2019

  10. [10]

    Value of information in feedback control: Quantification,

    T. Soleymani, J. S. Baras, and S. Hirche, “Value of information in feedback control: Quantification,”IEEE Trans. Autom. Control, vol. 67, no. 7, pp. 3730–3737, Jul. 2022

    work page 2022

  11. [11]

    On the value of information and mean squared error for noisy gaussian models,

    Z. Wang, M.-A. Badiu, and J. P. Coon, “On the value of information and mean squared error for noisy gaussian models,”IEEE Commun. Lett., vol. 26, no. 9, pp. 2023–2026, Sept. 2022

    work page 2023

  12. [12]

    A dynamic programming frame- work for vehicular task offloading with successive action improvement,

    Q. Li, Y . Hong, B. Lv, and R. Wang, “A dynamic programming frame- work for vehicular task offloading with successive action improvement,” IEEE Trans. Commun., vol. 73, no. 12, pp. 14 048–14 062, Dec. 2025

    work page 2025

  13. [13]

    Lyapunov-guided deep reinforcement learning for semantic-aware AoI minimization in UA V-assisted wireless networks,

    Y . Long, S. Gong, S. Sun, G. C. F. Lee, L. Li, and D. Niyato, “Lyapunov-guided deep reinforcement learning for semantic-aware AoI minimization in UA V-assisted wireless networks,”IEEE Trans. Wireless Commun., vol. 24, no. 8, pp. 6351–6364, Aug. 2025

    work page 2025

  14. [14]

    Enhancing energy efficiency in wireless-powered MEC systems through lyapunov-guided deep reinforcement learning,

    B. Zhu, L. Huang, K. Chi, A. Alharbi, K. Yu, and M. Guizani, “Enhancing energy efficiency in wireless-powered MEC systems through lyapunov-guided deep reinforcement learning,”IEEE Trans. Wireless Commun., vol. 24, no. 9, pp. 7563–7580, Sept. 2025

    work page 2025

  15. [15]

    Model predictive control enabled UA V trajectory optimization and secure resource allocation,

    Z. Li, C. Su, Z. Su, H. Peng, Y . Wang, W. Chen, and Q. Wu, “Model predictive control enabled UA V trajectory optimization and secure resource allocation,”IEEE Trans. Commun., vol. 73, no. 11, pp. 12 652–12 665, Nov. 2025

    work page 2025

  16. [16]

    IRS-enhanced secure semantic communication networks: Cross-layer and context-awared re- source allocation,

    L. Wang, W. Wu, F. Zhou, Z. Qin, and Q. Wu, “IRS-enhanced secure semantic communication networks: Cross-layer and context-awared re- source allocation,”IEEE Trans. Wireless Commun., vol. 24, no. 1, pp. 494–508, Jan. 2025

    work page 2025

  17. [17]

    Joint beam- forming in RIS-assisted multi-user transmission design: A model-driven deep reinforcement learning framework,

    W. Jin, J. Zhang, C.-K. Wen, S. Jin, and F.-C. Zheng, “Joint beam- forming in RIS-assisted multi-user transmission design: A model-driven deep reinforcement learning framework,”IEEE Trans. Commun., vol. 73, no. 5, pp. 3184–3198, May 2025

    work page 2025

  18. [18]

    Energy-efficient resource allocation for multi-user semantic communications: A deep reinforcement learning approach,

    X. Chen, J. Xu, W. Ni, S. Hu, Z. Qin, and S. Zhang, “Energy-efficient resource allocation for multi-user semantic communications: A deep reinforcement learning approach,”IEEE Wireless Commun. Lett., vol. 14, no. 5, pp. 1541–1545, May 2025

    work page 2025

  19. [19]

    AoI-aware resource allocation for platoon-based C-V2X networks via multi-agent multi-task reinforcement learning,

    M. Parvini, M. R. Javan, N. Mokari, B. Abbasi, and E. A. Jorswieck, “AoI-aware resource allocation for platoon-based C-V2X networks via multi-agent multi-task reinforcement learning,”IEEE Trans. Veh. Technol., vol. 72, no. 8, pp. 9880–9896, Aug. 2023

    work page 2023

  20. [20]

    Semantic communications for closed-loop physical AI systems,

    L. Wang, T. Shui, W. Saad, and P. Adjakple, “Semantic communications for closed-loop physical AI systems,” submitted toProc. IEEE Global Commun. Conf. (GLOBECOM), 2026

    work page 2026

  21. [21]

    A multi-task oriented semantic communication framework for autonomous vehicles,

    E. Eldeeb, M. Shehab, and H. Alves, “A multi-task oriented semantic communication framework for autonomous vehicles,”IEEE Wireless Commun. Lett., vol. 13, no. 12, pp. 3469–3473, Dec. 2024

    work page 2024

  22. [22]

    Robust semantic communications for speech transmission,

    Z. Weng, Z. Qin, and G. Y . Li, “Robust semantic communications for speech transmission,” inProc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP), Hyderabad, India, Apr. 2025, pp. 1–5

    work page 2025

  23. [23]

    Task-oriented low-label semantic communication with self-supervised learning,

    R. Gu, W. Xu, Z. Yang, D. Niyato, and A. Yener, “Task-oriented low-label semantic communication with self-supervised learning,”IEEE Trans. Wireless Commun., vol. 24, no. 11, pp. 9629–9644, Nov. 2025

    work page 2025

  24. [24]

    Channel- aware deep joint source-channel coding for multi-task oriented semantic communication,

    B. Wang, R. Gu, W. Xu, F. Jiang, M. Li, and S. Wang, “Channel- aware deep joint source-channel coding for multi-task oriented semantic communication,”IEEE Wireless Commun. Lett., vol. 14, no. 5, pp. 1521– 1525, May 2025

    work page 2025

  25. [25]

    Goal-oriented semantic communication for wireless visual question answering,

    S. Liu, N. Li, Y . Deng, and T. Q. S. Quek, “Goal-oriented semantic communication for wireless visual question answering,”IEEE J. Sel. Areas Commun., vol. 43, no. 12, pp. 4247–4261, Dec. 2025

    work page 2025

  26. [26]

    Importance of semantic information based on semantic value,

    S. Gao, X. Qin, L. Chen, Y . Chen, K. Han, and P. Zhang, “Importance of semantic information based on semantic value,”IEEE Trans. Commun., vol. 72, no. 9, pp. 5443–5457, Sept. 2024

    work page 2024

  27. [27]

    Digital twin of industrial networked control system based on value of information,

    V .-P. Bui, D. Abode, P. M. de Sant Ana, K. Muthineni, S. R. Pandey, and P. Popovski, “Digital twin of industrial networked control system based on value of information,” inProc. IEEE Global Commun. Conf. (GLOBECOM), Cape Town, South Africa, Dec. 2024, pp. 770–775

    work page 2024

  28. [28]

    V oi-driven joint optimization of control and communication in vehicular digital twin network,

    L. Lei, K. Zheng, J. Mei, and X. Shen, “V oi-driven joint optimization of control and communication in vehicular digital twin network,”IEEE Netw., vol. 39, no. 5, pp. 155–164, Sept. 2025

    work page 2025

  29. [29]

    DMWM: Dual- mind world model with long-term imagination,

    L. Wang, R. Shelim, W. Saad, and N. Ramakrishnan, “DMWM: Dual- mind world model with long-term imagination,” inProc. Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 38, San Diego, CA, USA, Dec. 2025, pp. 4911–4945

    work page 2025

  30. [30]

    Simplifying latent dynamics with softly state-invariant world models,

    T. Saanum, P. Dayan, and E. Schulz, “Simplifying latent dynamics with softly state-invariant world models,” inProc. Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 37, Vancouver, Canada, Dec. 2024, pp. 38 355– 38 382

    work page 2024

  31. [31]

    Dream to manipulate: Compositional world models em- powering robot imitation learning with imagination,

    L. Barcellona, A. Zadaianchuk, D. Allegro, S. Papa, S. Ghidoni, and E. Gavves, “Dream to manipulate: Compositional world models em- powering robot imitation learning with imagination,” inProc. Int. Conf. Learn. Represent. (ICLR), Singapore, Apr. 2025

    work page 2025

  32. [32]

    Learning latent dynamics for planning from pixels,

    D. Hafner, T. Lillicrap, I. Fischer, R. Villegas, D. Ha, H. Lee, and J. Davidson, “Learning latent dynamics for planning from pixels,” in International conference on machine learning (ICML), Long Beach, CA, USA, Jun. 2019, pp. 2555–2565

    work page 2019

  33. [33]

    Mastering diverse control tasks through world models,

    D. Hafner, J. Pasukonis, J. Ba, and T. Lillicrap, “Mastering diverse control tasks through world models,”Nature, vol. 640, no. 8059, pp. 647–653, Apr. 2025

    work page 2025

  34. [34]

    World model-based perception for visual legged locomotion,

    H. Lai, J. Cao, J. Xu, H. Wu, Y . Lin, T. Kong, Y . Yu, and W. Zhang, “World model-based perception for visual legged locomotion,” inProc. IEEE Int. Conf. Robot. Autom. (ICRA), Atlanta, GA, USA, May 2025, pp. 11 531–11 537

    work page 2025

  35. [35]

    World model- based learning for long-term age of information minimization in vehic- ular networks,

    L. Wang, R. Shelim, W. Saad, and N. Ramakrishnan, “World model- based learning for long-term age of information minimization in vehic- ular networks,”arXiv preprint arXiv:2505.01712, 2025

    work page arXiv 2025

  36. [36]

    Dual-mind world models: A general framework for learning in dynamic wireless networks,

    ——, “Dual-mind world models: A general framework for learning in dynamic wireless networks,”arXiv preprint arXiv:2510.24546, 2025

    work page arXiv 2025

  37. [37]

    Networld: Communication-based diffusion world model for multi-agent reinforce- ment learning in wireless networks,

    K. Meng, R. Li, Y . Deng, Z. Zhao, and H. Zhang, “Networld: Communication-based diffusion world model for multi-agent reinforce- ment learning in wireless networks,”arXiv preprint arXiv:2602.00558, 2026

    work page arXiv 2026

  38. [38]

    Mobiworld: World models for mobile wireless network,

    H. Chai, Y . Yuan, and Y . Li, “Mobiworld: World models for mobile wireless network,”arXiv preprint arXiv:2507.09462, 2025

    work page arXiv 2025

  39. [39]

    World models for cognitive agents: Transforming edge intelligence in future networks,

    C. Zhao, R. Zhang, J. Wang, G. Zhao, D. Niyato, G. Sun, S. Mao, and D. I. Kim, “World models for cognitive agents: Transforming edge intelligence in future networks,”arXiv preprint arXiv:2506.00417, 2025

    work page arXiv 2025

  40. [40]

    Td-mpc2: Scalable, robust world models for continuous control,

    N. Hansen, H. Su, and X. Wang, “Td-mpc2: Scalable, robust world models for continuous control,” inInternational Conference on Learning Representations (ICLR), Vienna, Austria, May 2024

    work page 2024

  41. [41]

    A theoretically- grounded codebook for digital semantic communications,

    L. Wang, R. Shelim, W. Saad, and N. Ramakrishnan, “A theoretically- grounded codebook for digital semantic communications,” inProc. IEEE Consum. Commun. Netw. Conf. (CCNC), Las Vegas, NV , USA, Jan. 2026, pp. 1–6

    work page 2026

  42. [42]

    Airsim: High-fidelity visual and physical simulation for autonomous vehicles,

    S. Shah, D. Dey, C. Lovett, and A. Kapoor, “Airsim: High-fidelity visual and physical simulation for autonomous vehicles,” inField and Service Robotics, Zurich, Switzerland, Sept. 2017, pp. 621–635

    work page 2017

  43. [43]

    Hoydis, S

    J. Hoydis, S. Cammerer, F. Ait Aoudia, M. Nimier-David, L. Maggi, G. Marcus, A. Vem, and A. Keller, “Sionna,” 2022, https://nvlabs.github.io/sionna/

    work page 2022