using, by a processor, a plurality of kinematic models to generate forecasted trajectories of an actor, the plurality of kinematic models being respectively associated with different actor types that are assigned to an actor detected in an environment of the vehicle;

obtaining, by the processor, levels of divergence for the forecasted trajectories from reference paths;

selecting, by the processor, a first kinematic model from the plurality of kinematic models based on the forecasted trajectories and a kinematic state of the actor, wherein a level of divergence associated with the forecasted trajectory generated using the first kinematic model is smallest one of the levels of divergences;

using, by the processor, the first kinematic model to predict a first path for the actor;

selecting, by the processor, a second kinematic model from the plurality of kinematic models responsive to movement of the actor no longer being consistent with typical movement of an object of one of the different actor types that is associated with the first kinematic model;

using, by the processor, the second kinematic model to predict a second path for the actor; and

controlling, by the processor, operations of the vehicle based on the first and second paths.