Retinal membrane peeling requires delicate manipulation. The presence of the surgeon's physiological tremor, the high variability and often low quality of the ophthalmic image, and excessive forces make the tasks more challenging. Preventing unintended movement caused by tremor and unintentional forces can reduce membrane injury. With the use of an actively stabilized handheld robot, we employ a monocular camera-based surface reconstruction method to estimate the retinal plane and we propose the use of a virtual fixture with application of hard and soft stops and motion scaling to improve control of the tool tip during delaminating in a laboratory simulation of retinal membrane peeling. A hard stop just below the membrane surface helps to limit downward force exerted on the surface. Motion scaling also improves the user's control of contact force when delaminating. We demonstrate a reduction of maximum force and maximum surface-penetration distance from the estimated retinal plane using the proposed technique.