RepresentationWorld models begin by constructing useful internal states from observations, sensors, language, video or other modalities.
DynamicsThey learn how states evolve through time and how actions can change the environment.
PredictionFuture states, observations, trajectories, rewards or outcomes can be estimated before they occur.
SimulationPossible futures can be generated or imagined without executing every action in the real world.
PlanningAgents can compare possible trajectories and select actions that better align with a goal.
ActionRobots, software agents and autonomous systems close the loop by acting and observing the result.