Research on radio-frequency (RF) sensing for augmented reality (AR) has produced a variety of prototypes—from
RF-driven casualty triage to threat detection—but the lack
of standardized datasets and evaluation frameworks hampers
reproducibility. Machine learning workflows are often fragmented and informal; datasets, code and configurations are
loosely coupled, making it difficult to trace experiments and
reproduce results [1]. Reproducibility requires capturing not just
data and code, but also the process and decisions behind an
experiment [2]. We introduce OpenBench-AR, an open-source
benchmark suite that provides standardized RF traces, JSON
metrics and LaTeX figure/table autogeneration for RF-to-AR
systems. OpenBench-AR packages a client simulator, exporters
and a README.md that guide users through dataset reproduction
and one-command generation of evaluation figures. We demonstrate OpenBench-AR on prior RF-AR pipelines, showing how
researchers can reproduce latency, frame rate and power results
across hardware. Our artifact is ready for artifact evaluation
tracks at ReproNLP/MLSys and systems demo workshops.