Ilmu Komputer & AI editorial
Unified Evaluation Methodology for AI-Native Integrated Sensing and Communication
The core problem
Integrated Sensing and Communication (ISAC) couples radio sensing, data transmission, and control actions within a single closed-loop system. When Artificial Intelligence (AI)-driven policies adapt sensing and communication online across a variety of sensing tasks and objectives, end-to-end performance is shaped not only by waveform and channel conditions but also by inference latency, uncertainty, environmental dynamics, and hardware non-idealities. This leads to fundamental trade-offs between sensing accuracy, communication reliability, and resource overhead.
The manuscript addresses a critical gap: theoretical ISAC gains often fail to translate into deployment-ready performance claims because evaluation methodologies are fragmented across heterogeneous testbeds, metrics, and assumptions. The authors define **AI-native ISAC** as ISAC in which learning-based agents adapt sensing, communication, and actuation policies online under uncertainty. The central contribution is a unified system architecture and evaluation methodology that formalizes the design space of closed-loop ISAC, proposes a three-stage validation pipeline, and provides a minimal reporting checklist linking technic
Innovation
The manuscript presents the unified architecture and evaluation methodology as a conceptual and procedural contribution rather than reporting experimental results from a single deployment. The results are therefore the formalized design space, the three-stage validation pipeline, and the minimal reporting checklist.
Key results include:
- A formal definition of AI-native ISAC as ISAC with online adaptation of sensing, communication, and actuation policies under uncertainty.
- A three-stage validation pipeline that progresses from bounds and feasibility analysis, through high-fidelity digital-twin simulation, to preliminary over-the-air validation.
- A minimal reporting checklist that links technical KPIs (data rate, SINR, target detection, parameter estimation, track quality, localization error, outage, latency, overhead, energy per decision) to application-level KVIs (availability, mission effectiveness).
- Two representative instantiations (UAV-based outdoor and RIS-enabled indoor coverage) that illustrate how to structure reproducible baselines and comparable evidence across heterogeneous deployments.
The authors emphasize that the pipeline helps bridge the gap between theoret
Why it matters
The manuscript makes a significant contribution by addressing the evaluation gap in AI-native ISAC. The proposed three-stage validation pipeline is a structured approach that balances theoretical rigor, simulation fidelity, and real-world validation. The minimal reporting checklist is a practical tool that can be adopted by researchers and practitioners to ensure reproducibility and comparability.
The formalization of the design space using a closed-loop control perspective is useful for identifying trade-offs. The equations presented in the Methodology section capture the essential dynamics: the policy selects actions based on observations, the environment evolves according to , and observations are generated by . The uncertainty terms and explicitly represent environmental dynamics and hardware non-idealities, which are often ignored in simplified evaluations.
The two representative instantiations highlight the diversity of ISAC deployments. UAV-based outdoor ISAC must contend with mobility, 3D channel effects, and airborne hardware constraints. RIS-enabled indoor coverage must model reconfigurable surface configurations, indoor propagation, and blockage dynamics. The unified methodology allows these heterogeneous systems to be evaluated using a common framework, which is a step toward standardized benchmarking.
One limitation is that the manuscript does not provide empirical results from applying the pipeline to the two instantiations. Future work could involve implementing the pipeline and reporting quantitative results. Another limitation is that the reporting checklist may need to be extended for emerging ISAC applications, such as vehicular networks or industrial IoT.
The authors' emphasis on linking KPIs to KVIs is particularly important. Technical metrics alone do not capture whether a system is useful for a given mission. By requiring application-level KVIs such as availability and mission effectiveness, the methodology ensures that evaluations are relevant to end users.
Overall, the manuscript provides a timely and necessary framework for evaluating AI-native ISAC. As the field moves from theory to deployment, unified evaluation methodologies will be essential for comparing approaches, identifying best practices, and accelerating progress.
Who should read this
Opening member content…