Abstract: Transcranial magnetic stimulation (TMS) is a neuro-modulation method. In clinical practice, the placement and orientation of TMS coils are manually determined by doctors based on their experience, resulting in inaccurate coil positioning and poor repeatability. To address these issues, this paper proposed a TMS coil robot-assisted positioning system that utilized an RGB camera instead of dual infrared cameras in the navigation system. Additionally, a markerless TMS coil robot-assisted positioning method based on neural networks was introduced. A neural network was built to map the coil attitude of camera space to the joint angle of manipulator space and trained using simulation data to verify that the neural network architecture was suitable for TMS coil pose placement. Subsequently, experimental verification confirmed the feasibility of this method while demonstrating that the trained neural network had good generalization ability in the task of TMS coil localization. Finally, validation results in Cartesian space show an average position error of 2.16 mm for the TMS coil's three-dimensional location and an overall attitude error of 0.055 rad, indicating that the proposed TMS coil robot-assisted positioning system using RGB camera achieves similar accuracy levels as other research or commercial systems employing dual infrared cameras, meeting clinical requirements for TMS treatment and demonstrating its feasibility for clinical applications.