Skip to main content

EnvData

class EnvData(BaseModel)

RL-specific data class built by RLExecutor and passed to RLAgentEnv constructors. The core ADK does not construct EnvData; custom executors use OptimizationContext instead. Built-in and custom environments on the RL path read from env_data to configure spaces and to call back into the simulator.

Import​

from adk.executors.rl import EnvData

Definition​

class EnvData(BaseModel):
    static_world_controls_space: gym.spaces.Dict
    optimized_world_controls_space: gym.spaces.Dict
    randomized_world_controls_space: gym.spaces.Dict

    default_static_world_controls: dict[str, DesignParamValue]
    default_optimized_world_controls: dict[str, DesignParamValue]
    default_randomized_world_controls: dict[str, DesignParamValue]

    world_observations_space: gym.spaces.Dict
    default_world_observations: dict[str, ObservationValue]

    internal_structure_graph: nx.Graph | None
    default_world_features: dict[str, dict[str, float]] | None

    targets_space: gym.spaces.Dict

    step_world: StepWorldCB

    optimization_data: OptimizationData

Members​

World control spaces​

static_world_controls_space, optimized_world_controls_space, and randomized_world_controls_space are Gymnasium Dict spaces describing controllable parameters that are fixed, optimized, or randomly sampled.

Default controls​

default_static_world_controls, default_optimized_world_controls, and default_randomized_world_controls hold starting values for each group.

Observations and targets​

world_observations_space and default_world_observations describe evaluated measurements from the simulator. targets_space describes target value ranges for the optimization.

Graph and features​

internal_structure_graph is present when the model uses graph instrumentation. default_world_features holds optional internal feature values at default parameters.

step_world​

step_world: StepWorldCB

Callback that applies design parameters to the simulator and returns raw observations, evaluated observations, and optional features. Used by OptimizationEnv implementations.

optimization_data​

optimization_data: OptimizationData

Contains the inference flag and loaded GenieModel for this run.

See also RL Run Data.