Service robots are increasingly applied in real life environments populated with human beings. In such a challenging scenario, the autonomous robots have to avoid collision in a “natural” way, that is to execute trajectories that a human would follow. This challenging goal can be efficiently tackled if a sufficiently descriptive human motion model is available, in order to predict future pedestrian behaviour and hence safely planning the correct route. In this paper, we move a first step towards a motion model that is able to describe to a certain extent the nonverbal negotiation of spaces in shared environments, still preserving its simplicity for ease of computation. The avoidance task is shared among the robot and the pedestrians and thus human-like trajectories can be generated. Simulations and application to actual pedestrian data are presented to validate the model.