The event segmentation theory (EST) postulates that humans systematically segment the continuous sensorimotor information flow into events and event boundaries. The basis for the observed segmentation tendencies, however, remains largely unknown. We introduce a computational model that grounds EST in the interaction abilities of a system. The model learns events and event boundaries based on actively gathered sensorimotor signals. It segments the signals based on principles of probabilistic predictive coding and surprise. The implemented model essentially simulates, anticipates, and learns event progressions and event transitions online while interacting with the environment by means of dynamic, predictive Bayesian models. Besides the model’s event segmentation capabilities, we show that the learned encodings can be used for higher-order planning. Moreover, the encodings systematically conceptualize environmental interactions and they help to identify the factors that are critical for ensuring interaction success.