Artificial central pattern generators (CPGs) framework is well adapted to the control of bio-mimetic systems during rhythmic tasks like locomotion. They have the ability to reproduce biological behavior as well as to be used as feedforward controllers for multi-articulated systems. In this paper we present a model of human gait activity based on an oscillator network. The model is especially dedicated to reproduce trunk muscular activities as observed in previous studies, and to fill a lack in trunk modeling in human gait simulation. An offline validation is performed using recorded accelerometer signal that monitors trunk movements during locomotion. We are able to control the model based on this real data and to adapt its pattern to contextual changes (stairs, slope).