An action probability factor is developed based at least in part on a plurality of probability arrays predicting one or more probabilities of a deviation from at least one of a planned vehicle direction, position, speed, and acceleration. Levels of autonomous control are transitioned based at least in part on the action probability factor.