<b><i>Background/Purpose:</i></b> Few data are available on the learning curve (LC) in robot-assisted pancreaticoduodenectomy (RAPD) and no study specifically addresses the LC of a single surgeon. <b><i>Methods:</i></b> The LC of a single surgeon in RAPD was determined using the cumulative sum method, based on operative time (OT). Data were extracted from a prospectively maintained database and analyzed retrospectively considering all events occurring within 90 days of index operation. <b><i>Results:</i></b> Seventy RAPD were analyzed. One operation was converted to open surgery (1.4%). One patient died within 30 days (1.4%) and one within 90 days (2.8%). Postoperative complications occurred in 53 patients (75.7%) and exceeded Clavien-Dindo grade IIIb in 7 patients (10%). OT dropped after 33 operations from a mean of 564 ± 101.7 min to a mean of 484.1 ± 77.9 min (p = 0.0005) and was associated to reduced incidence of delayed gastric emptying (72.7 vs. 48.7%; p = 0.039). The rate of hospital readmission improved after 40 operations from 20.0 (8 of 40) to 3.3% (1 of 30) (p = 0.04). <b><i>Conclusions:</i></b> RAPD was safely feasible in selected patients. OT dropped after the first 33 operations and was associated with reduced rate of delayed gastric emptying. Readmission rate improved after 40 operations.