Struct WorldModel 

pub struct WorldModel<B: Backend, D: ActionConditionedPredictor<B>, C: CostFunction<B>> {
    pub dynamics: D,
    pub cost: C,
    /* private fields */
}

World model that can simulate trajectories and evaluate plans.

Combines a dynamics model and cost function for model-based planning. The encoder is external — state representations are passed in directly.

Fields

dynamics: D

Dynamics model: predicts next state given current state and action.

cost: C

Cost function: evaluates how close a trajectory gets to the goal.

Implementations

impl<B: Backend, D: ActionConditionedPredictor<B>, C: CostFunction<B>> WorldModel<B, D, C>

pub fn new(dynamics: D, cost: C) -> Self

Create a new world model.

pub fn rollout( &self, initial_state: &Representation<B>, actions: &[Action<B>], ) -> Vec<Representation<B>>

Simulate a sequence of actions starting from an initial state.

Returns the full trajectory including the initial state.

Arguments

• initial_state - Starting state representation
• actions - Sequence of actions to simulate

Returns

Trajectory of len(actions) + 1 states (initial + predicted)

pub fn evaluate_plan( &self, initial_state: &Representation<B>, actions: &[Action<B>], goal: &Representation<B>, ) -> Energy<B>

Evaluate a plan by computing its total cost relative to a goal.

Arguments

• initial_state - Starting state
• actions - Sequence of actions (the plan)
• goal - Goal state to reach

Returns

Cost of the plan (lower = better)