Experimental validation of the structured model-based framework

Test experimentally whether the structured model-based framework combining a physics-informed surrogate, gradient-based sensitivities, and constrained optimization transfers more reliably across tokamak devices than end-to-end learned controllers.

Background

The paper hypothesizes that incorporating physical structure, differentiable surrogate sensitivities, and explicit actuator constraints will improve transferability relative to end-to-end learned control. The study provides only offline ARC V3A simulations, leaving this cross-device transfer claim unresolved.

References

We expect the structured model-based framework physics-informed surrogate, gradient-based sensitivities, constrained optimization to transfer more reliably across devices than end-to-end learned controllers, though this hypothesis remains to be tested experimentally.

— On the feasibility of model-based feedback control of vertical instability growth rate using out-vessel coils in ARC-like scenarios  (2608.22677 - Kumar et al., 24 Aug 2026) in Section 6, Conclusions