Concepts
The layers and mechanisms connecting world models with language, visual perception, planning, and embodied action.
Reading path
Large Language Model
Models statistical structure in language and produces or interprets text.
PerceptionVision-Language Model
Connects visual observations with language representations and instructions.
SimulationWorld model
Represents how an environment may change, including the consequences of actions.
PlanningPlanner
Searches possible actions or trajectories against a goal and predicted outcomes.
AgencyAgent
Observes, chooses, acts, and updates its behavior across a feedback loop.
EmbodimentPhysical AI
Applies learned perception, prediction, and control in embodied systems.
SystemsDigital twin
Maintains an operational digital representation linked to a physical system.
Supporting mechanisms
Latent state
A compressed internal representation used to predict relevant environment dynamics.
PlanningAction conditioning
Conditions a prediction on a proposed control or intervention.
SimulationUncertainty
Represents what a model does not know or which futures remain plausible.
EmbodimentSim-to-real
Transfers behavior learned or tested in simulation into a physical setting.
SimulationInteractive control
A prediction loop that lets a user steer the simulated environment frame by frame.
FoundationSelf-supervised pretraining
Learns representations from unlabeled video, text, or sensor streams before downstream adaptation.