pith. sign in

arxiv: 2604.19599 · v1 · submitted 2026-04-21 · 📡 eess.SP · cs.NI

Active Inference-Enabled Agentic Closed-Loop ISAC with Long-Horizon Planning

Guangjin Pan , Zhuojun Tian , Mehdi Bennis , Henk Wymeersch This is my paper

Pith reviewed 2026-05-10 01:50 UTC · model grok-4.3

classification 📡 eess.SP cs.NI
keywords active inferenceintegrated sensing and communicationclosed-loop systemsdigital twinresource allocationagentic systemsfactor graphlocalization
0
0 comments X

The pith

An active inference agent enables closed-loop ISAC by jointly optimizing control and sensing resource allocation through factor graph message passing on a maintained digital twin model.

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

This paper establishes that active inference can drive wireless agents to jointly optimize sensing and control in closed-loop integrated sensing and communication systems. The agent maintains a generative model as a digital twin that combines a localization model for uncertainty-aware state inference with a localization channel knowledge map to approximate observation quality. It performs backward-forward message passing on a factor graph to plan actions over long horizons. Simulations show the approach adapts sensing resources to spatially varying channel conditions and achieves a better balance among tracking accuracy, control effort, and sensing resource consumption than baseline strategies.

Core claim

The AIF-enabled agent adaptively allocates sensing resources based on spatially varying channel conditions by performing backward-forward message passing on a factor graph, achieving a superior balance among tracking accuracy, control effort, and sensing resource consumption over baseline strategies.

What carries the argument

Backward-forward message passing on a factor graph for active inference planning, supported by a generative model maintained as a digital twin that includes a localization model and localization channel knowledge map.

Load-bearing premise

The generative model maintained as a digital twin, including the localization model and localization channel knowledge map, sufficiently approximates real-world dynamics and observation quality to enable effective long-horizon planning.

What would settle it

A physical experiment where actual channel conditions and observation quality deviate from the localization channel knowledge map predictions, causing the agent's performance to fall below that of baseline strategies.

Figures

Figures reproduced from arXiv: 2604.19599 by Guangjin Pan, Henk Wymeersch, Mehdi Bennis, Zhuojun Tian.

Figure 1
Figure 1. Figure 1: AIF-enabled wireless agentic system for closed-loop ISAC. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Factor graph for the planning stage: (a) backward pass for computing [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: AIF trajectory colored by the selected k, overlaid on the observa￾tion variance map. AIF allocates more subcarriers in high-variance regions (Zone 1) and fewer in low-variance regions (Zone 2). overlapping splits to prevent overfitting. Both models are trained with the Adam optimizer at a learning rate of 10−4 for 200 epochs. For the closed-loop simulation, the ME follows a straight-line reference trajecto… view at source ↗
read the original abstract

Wireless agentic systems enable agents to autonomously perceive, reason, and act. However, existing works neglect the tight coupling between sensing and control in closed-loop integrated sensing and communication (ISAC) systems. In this paper, we propose an active inference (AIF)-driven wireless agentic system for closed-loop ISAC, which jointly optimizes control and sensing resource allocation via backward--forward message passing on a factor graph. The AIF agent maintains a generative model as a digital twin by integrating a localization model for uncertainty-aware state inference and a localization channel knowledge map (CKM) for approximating observation quality during planning. Simulation results demonstrate that the AIF-enabled agent adaptively allocates sensing resources based on spatially varying channel conditions, achieving superior balance among tracking accuracy, control effort, and sensing resource consumption over baseline strategies.

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

1 major / 1 minor

Summary. The paper proposes an active inference (AIF)-driven agentic system for closed-loop integrated sensing and communication (ISAC). The AIF agent maintains a generative model as a digital twin by integrating a localization model for uncertainty-aware state inference and a localization channel knowledge map (CKM) for approximating observation quality. It performs long-horizon planning via backward-forward message passing on a factor graph to jointly optimize control and sensing resource allocation. Simulation results show that the agent adaptively allocates sensing resources based on spatially varying channel conditions, achieving a superior balance among tracking accuracy, control effort, and sensing resource consumption compared to baseline strategies.

Significance. If the digital twin approximation holds under realistic conditions, the work could advance agentic wireless systems by providing a principled framework for uncertainty-aware, long-horizon joint sensing-control optimization using active inference and factor graphs. The approach credits the use of message passing for planning and the integration of CKM for observation quality modeling. However, significance is limited by the simulation-based evidence without demonstrated robustness to model mismatch.

major comments (1)
  1. [Simulation results] Simulation results section: The central claim of superior adaptive resource allocation and balanced performance rests on simulations that appear to assume perfect statistical match between the digital twin (localization model + CKM) and the environment generating observations. No experiments introduce mismatches such as unmodeled spatial correlations, hardware noise, or time-varying CKM errors. This directly undermines the weakest assumption that the generative model sufficiently approximates real-world dynamics for effective closed-loop planning, as the reported gains do not probe the failure regime.
minor comments (1)
  1. [Abstract] The abstract states performance superiority without providing quantitative metrics, baseline definitions, or error measures; ensure the full results section includes these for verifiability.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. The major comment raises a valid point about the scope of our simulation evaluation, which we address below with a commitment to revision.

read point-by-point responses

  1. Referee: Simulation results section: The central claim of superior adaptive resource allocation and balanced performance rests on simulations that appear to assume perfect statistical match between the digital twin (localization model + CKM) and the environment generating observations. No experiments introduce mismatches such as unmodeled spatial correlations, hardware noise, or time-varying CKM errors. This directly undermines the weakest assumption that the generative model sufficiently approximates real-world dynamics for effective closed-loop planning, as the reported gains do not probe the failure regime.

    Authors: We agree that the current simulations assume a perfect statistical match between the digital twin (localization model and CKM) and the true environment, which is a standard initial step to isolate and demonstrate the core benefits of the AIF-driven planning. This does represent a limitation, as it leaves untested the framework's behavior under realistic model mismatches. In the revised manuscript, we will add a new set of experiments that deliberately introduce controlled mismatches, including unmodeled spatial correlations in the channel, additive hardware noise, and time-varying perturbations to the CKM. These results will quantify performance degradation and show how the long-horizon factor-graph planning adapts (or fails to adapt) in the failure regime, thereby strengthening the evidence for practical applicability. revision: yes

Circularity Check

0 steps flagged

No circularity identified in the proposed AIF-ISAC framework

full rationale

The paper proposes an active inference agent for closed-loop ISAC that performs joint optimization of control and sensing via backward-forward message passing on a factor graph, while maintaining a generative model (localization model plus localization CKM) as a digital twin. Performance is evaluated through simulation results that compare adaptive resource allocation against baselines. No equations or steps in the described derivation chain reduce by construction to their own inputs; the digital twin and CKM are introduced as explicit modeling choices rather than outputs fitted from the target results. Any self-citations are not load-bearing for the central claim, and the simulation-based evidence remains independent of the framework itself under the stated assumptions.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the assumption that the generative model and CKM provide sufficiently accurate approximations for planning; no free parameters or invented entities are explicitly quantified in the abstract.

axioms (1)
  • domain assumption The generative model as a digital twin accurately captures localization uncertainty and observation quality for planning purposes.
    Invoked to justify the use of the model for adaptive resource allocation and long-horizon planning.
invented entities (1)
  • Localization channel knowledge map (CKM) no independent evidence
    purpose: Approximating observation quality during planning within the digital twin
    Introduced as a component of the generative model to enable uncertainty-aware planning.

pith-pipeline@v0.9.0 · 5444 in / 1377 out tokens · 148912 ms · 2026-05-10T01:50:41.255725+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

15 extracted references · 15 canonical work pages

  1. [1]

    Advanced architectures integrated with agentic AI for next-generation wireless networks,

    K. Devet al., “Advanced architectures integrated with agentic AI for next-generation wireless networks,”IEEE Communications Standards Magazine, 2025

    work page 2025

  2. [2]

    Semantic communication for rate-limited closed- loop distributed communication-sensing-control systems,

    G. Panet al., “Semantic communication for rate-limited closed- loop distributed communication-sensing-control systems,”arXiv preprint arXiv:2512.19177, 2025

    work page arXiv 2025

  3. [3]

    Active inference framework for closed-loop sens- ing, communication, and control in UA V systems,

    G. Panet al., “Active inference framework for closed-loop sens- ing, communication, and control in UA V systems,”arXiv preprint arXiv:2509.14201, 2025

    work page arXiv 2025

  4. [4]

    Networking and communications in autonomous driving: A survey,

    J. Wanget al., “Networking and communications in autonomous driving: A survey,”IEEE Commun. Surveys Tuts., vol. 21, no. 2, pp. 1243–1274, 2019

    work page 2019

  5. [5]

    Goal-oriented semantic communications for robotic waypoint transmission: The value and age of information approach,

    W. Wuet al., “Goal-oriented semantic communications for robotic waypoint transmission: The value and age of information approach,” IEEE Trans. Wireless Commun., vol. 23, no. 12, pp. 18 903–18 915, 2024

    work page 2024

  6. [6]

    Maximizing the value of service provisioning in multi- user ISAC systems through fairness guaranteed collaborative resource allocation,

    B. Liet al., “Maximizing the value of service provisioning in multi- user ISAC systems through fairness guaranteed collaborative resource allocation,”IEEE J. Sel. Areas Commun., vol. 42, no. 9, pp. 2243–2258, 2024

    work page 2024

  7. [7]

    Efficient uav hovering, resource allocation, and trajectory design for ISAC with limited backhaul capacity,

    A. Khaliliet al., “Efficient uav hovering, resource allocation, and trajectory design for ISAC with limited backhaul capacity,”IEEE Trans. Wireless Commun., vol. 23, no. 11, pp. 17 635–17 650, 2024

    work page 2024

  8. [8]

    Energy-efficient beamforming design for integrated sensing and communications systems,

    J. Zouet al., “Energy-efficient beamforming design for integrated sensing and communications systems,”IEEE Trans. Commun., vol. 72, no. 6, pp. 3766–3782, 2024

    work page 2024

  9. [9]

    Distributed cooperative LQR design for multi-input linear systems,

    P. Duanet al., “Distributed cooperative LQR design for multi-input linear systems,”IEEE Trans. Control Netw. Syst., vol. 10, no. 2, pp. 680–692, 2022

    work page 2022

  10. [10]

    Co-design of sensing, communications, and control for low-altitude wireless networks,

    H. Jinet al., “Co-design of sensing, communications, and control for low-altitude wireless networks,”IEEE Trans. Mobile Comput., no. 01, pp. 1–13, 2025

    work page 2025

  11. [11]

    A review of uncertainty for deep reinforcement learning,

    O. Lockwoodet al., “A review of uncertainty for deep reinforcement learning,” inProc. AAAI‘2022, vol. 18, no. 1, 2022, pp. 155–162

    work page 2022

  12. [12]

    The free-energy principle: a unified brain theory?

    K. Friston, “The free-energy principle: a unified brain theory?”Nature reviews neuroscience, vol. 11, no. 2, pp. 127–138, 2010

    work page 2010

  13. [13]

    AI-driven wireless positioning: Fundamentals, standards, state-of-the-art, and challenges,

    G. Panet al., “AI-driven wireless positioning: Fundamentals, standards, state-of-the-art, and challenges,”IEEE Commun. Surveys Tuts., vol. 28, pp. 4394–4428, 2026

    work page 2026

  14. [14]

    An uncertainty-aware performance measure for multi- object tracking,

    J. Pintoet al., “An uncertainty-aware performance measure for multi- object tracking,”IEEE Signal Processing Letters, vol. 28, pp. 1689– 1693, 2021

    work page 2021

  15. [15]

    “Sionna,” https://nvlabs.github.io/sionna/

    work page