Matched comparison with alternative predictive architectures

Conduct a matched comparison between the behavior–realization architecture and closely related reference-governor or predictive-safety-filter architectures, using a faithful implementation of the relevant alternative methods.

Background

The paper positions its predictive realization layer as architecturally similar to control barrier function quadratic programs, reference governors, and predictive safety filters because each modifies an upstream desired command or behavior to satisfy physical constraints. However, the present study does not implement or experimentally compare against these alternatives. The authors identify a matched comparison with the closest alternative architecture as an unresolved matter needed to assess the claimed architectural distinction against established approaches.

References

We do not compare against a CBF-QP filter or a reference-governor instance in this paper; Section~\ref{sec:scope} lists a matched comparison against the closest alternative architecture as open.

— Behavior--Realization Separation for Constrained Physical Human--Robot Interaction  (2609.00669 - Cao, 1 Sep 2026) in Section 2, Related Work and Positioning, paragraph discussing control barrier function quadratic programs

First, a shield driven by a current-state threshold rather than by a forecast: the reconfiguration transient lasts tens of milliseconds while the tank time constant is minutes, so a 10-second forecast horizon may add little over simply checking whether the process is currently near its setpoint, and until that baseline is run the LSTM's contribution over a threshold is not established.

— BALLAST: Physics-Aware Moving Target Defense for Industrial Control Systems Using a Digital Twin Safety Oracle  (2610.18311 - Mogollon-Gutierrez et al., 13 Oct 2026) in Section 5.7, Limitations and Future Directions, subsection “The contribution of each component is only partly isolated”

Third, a PPO-Lagrangian agent: we argue that a constraint satisfied in expectation is inadequate here, but we did not run that comparison and the argument remains analytical.

— BALLAST: Physics-Aware Moving Target Defense for Industrial Control Systems Using a Digital Twin Safety Oracle  (2610.18311 - Mogollon-Gutierrez et al., 13 Oct 2026) in Section 5.7, Limitations and Future Directions, subsection “The contribution of each component is only partly isolated”