Sir Isaac Newton famously remarked that doing science he was like "...a boy playing on the sea-shore ... finding a smoother pebble or a prettier shell than ordinary," vividly depicting how a scientist shifts the research focus along the career. Indeed, changing research interest is an integral part of a scientific career. Despite extensive investigations on various factors that influence a scientist's choice of research topics, little is known about macroscopic patterns characterizing research interest evolution undertaken by an ensemble of scientists. Here using publication records of over 14,000 authors in physics, we perform a quantitative analysis of the extent of research interest change, which demonstrates that most of scientists are characterized by little change. We identify three fundamental features that shape the research interest evolution. Using a simple random walk based model, we successfully reproduce the patterns observed in empirical data. The results fill a critical gap in our large-scale quantitative understanding of science and uncover previously unknown regularities underlying individual careers.