For its own safety, a robot system should never find itself in a state where there is no feasible trajectory to avoid collision with an obstacle. Such a state is an Inevitable Collision State (ICS). The ICS concept is particularly useful for navigation in dynamic environments because it takes into account the future behaviour of the moving objects. Accordingly it requires a model of the future evolution of the environment. In the real-world, the future trajectories of the obstacles are generally unknown and only estimates are available. This paper introduces a probabilistic formulation of the ICS concept which incorporates uncertainty in the model of the future trajectories of the obstacles. It also presents two novel probabilistic ICSchecking algorithms that are compared with their deterministic counterpart.