We propose an efficient approach for focal-plane wavefront estimation for high-contrast imaging. It is designed for very low photon flux regimes with the Roman Space Telescope Coronagraph Instrument in mind. During days-long exposures of Roman, the starlight wavefront will drift due to the thermal instabilities of the optical tube assembly (OTA) and deformable mirror (DM) actuators. To correct the drift in high-order spatial wavefront modes, Roman will periodically be pointed to a bright reference star to sense the speckles in the dark hole (the high-contrast region of the image) and correct them. However, this maneuver decreases observation time by several hours a day and may lead to wavefront instability by causing structural and thermal stresses on the OTA. Instead, we propose maintaining the dark hole using the DM to modulate (probe) the speckles while observing the target star. Our proposed algorithm can then be used to directly estimate the DM commands that would create such speckles, rather than first estimating the electric field of the speckles. Applying the opposite commands on the DM reduces the intensity of the speckles without slewing to the reference star. Further, we analyze this dark-hole maintenance approach using data from Roman Observing Scenario 9 (OS9) simulations while assuming that the high-order wavefront control on Roman will be carried out from the Earth (ground-in-the-loop). In our simulations of Roman’s Hybrid Lyot Coronagraph, the high-order wavefront errors were sensed and corrected while observing a dim target, 47 UMa. Despite the DM probes, the photon flux in the dark hole remained close to that in OS9 and well below detector noise. In terms of planet-detectability, dark hole maintenance kept the post-processing contrast within 10% of OS9’s, even after pointing the telescope at the target star for 6 days. The OTA stability that dark hole maintenance provides compared to slewing the telescope potentially outweighs the small negative impact it has on the contrast.