This thesis demonstrates autonomous single-molecule manipulation on a strongly interacting organic-metal interface by combining scanning probe microscopy with reinforcement learning. Using PTCDA adsorbed on Ag(111) as a model system, an autonomous agent learns to extract individual molecules from a densely packed monolayer, a task governed by pronounced intermolecular interactions and complex tip-molecule interactions. Systematic characterization of the tip-PTCDA junction based on 4,096 vertical two-contact manipulation experiments reveals the formation of chemically specific tip-oxygen bonds and pronounced flexibility of the molecular corners. Distinct responses of the two inequivalent carboxylic oxygen species indicate locally varying binding geometries and enable the identification of lateral “corridors” that favor peeling-like extraction trajectories. These trajectories were validated through both hand-controlled manipulation and autonomous manipulation. Following detachment from the surface, PTCDA molecules frequently remain vertically bound to the tip apex. Experiments employing an artificial surface-anchored apex demonstrate that a geometrically non-coordinated carboxylic oxygen atom experiences attractive interactions with the substrate, driving molecular reorientation and promoting the formation of a second tip-oxygen bond. Controlled quarter-circular trajectories reproducibly yield symmetric diagonal contact geometries, a behavior that is not observed upon lifting isolated adsorbed molecules. Building on these insights, reinforcement-learning agents were trained to optimize molecular extraction trajectories through trial and error and to adapt to variations in the tip apex configuration. Pretrained agents exhibit superior performance, highlighting the transferability of learned manipulation strategies across different experimental conditions. Finally, the enhanced manipulation capabilities are employed to fabricate a single-molecule probe for scanning quantum dot microscopy, enabling high-resolution mapping of the electrostatic potential above PTCDA/Ag(111) monolayers and molecular vacancies. Distinct charge-transfer behaviors associated with different molecular orientations are observed, providing new insights into electronic properties at the single-molecule scale. Together, these results pave the way toward autonomous fabrication and in situ characterization of complex functional nanostructures.
Philipp Kopp (Wed,) studied this question.