Verify the source of implicit excitation in closed-loop control
Verify whether future control actions selected by the improved cross-entropy method provide implicit excitation through their execution and incorporation into the rolling inference context, thereby explaining the strong closed-loop performance observed from initially unexcited contexts.
References
Surprisingly, even when the initial context at the start of an episode is unexcited, the performance is good. We hypothesize that this arises because future control actions selected by \ac{iCEM} are subsequently executed and enter the rolling context, thereby providing implicit excitation. In contrast, the data used as \ac{MPC} context in the open-loop experiments is generated by a deterministic \ac{MPC}, with no learning or inference dynamics. Verifying this mechanism directly is left for future work.
More broadly, our results suggest that context is a central design variable for \ac{TSFM}-based control: how to retain control-informative observations and how much excitation is required for reliable input-output identification remain open questions.