flatland.core.env_observation_builder module#
ObservationBuilder objects are objects that can be passed to environments designed for customizability. The ObservationBuilder-derived custom classes implement 2 functions, reset(env) and get() or get(handle).
reset(env) is called after each environment reset, to allow for pre-computing relevant data. It receives the (possibly newly generated) env instance, so any instantiations depending on env parameters (e.g. width, height) should be done here rather than in __init__.
get() is called whenever an observation has to be computed, potentially for each agent independently in case of multi-agent environments.
- class flatland.core.env_observation_builder.DummyObservationBuilder[source]#
Bases:
ObservationBuilder[Environment,bool]DummyObservationBuilder class which returns dummy observations This is used in the evaluation service
- get(handle: int = 0) bool[source]#
Called whenever an observation has to be computed for the env environment, possibly for each agent independently (agent id handle).
Parameters#
- handleint, optional
Handle of the agent for which to compute the observation vector.
Returns#
- function
An observation structure, specific to the corresponding environment.
- class flatland.core.env_observation_builder.ObservationBuilder[source]#
Bases:
Generic[EnvType,ObservationType]ObservationBuilder base class.
- get(handle: int = 0) ObservationType[source]#
Called whenever an observation has to be computed for the env environment, possibly for each agent independently (agent id handle).
Parameters#
- handleint, optional
Handle of the agent for which to compute the observation vector.
Returns#
- function
An observation structure, specific to the corresponding environment.
- get_many(handles: List[int] | None = None) Dict[int, ObservationType][source]#
Called whenever an observation has to be computed for the env environment, for each agent with handle in the handles list.
Parameters#
- handleslist of handles, optional
List with the handles of the agents for which to compute the observation vector.
Returns#
- function
A dictionary of observation structures, specific to the corresponding environment, with handles from handles as keys.
- reset(env: EnvType)[source]#
Called after each environment reset, to allow for pre-computing relevant data. Receives the (possibly newly generated) env instance, so any instantiations depending on env parameters (e.g. width, height) should be made here rather than in __init__.
Subclasses that need to pre-compute env-dependent data should override this method and call super().reset(env) first to keep self.env up to date.
Parameters#
- envEnvType
the (possibly newly generated) environment instance
- flatland.core.env_observation_builder.gauss_perturbation_observation_builder_wrapper(builder: ObservationBuilder[Environment, ndarray], np_random: RandomState, mu: ndarray | None = None, sigma: ndarray | None = None) ObservationBuilder[Environment, ndarray][source]#
Perturb a numpy array based observation with Gaussian noise.
Parameters#
builder : ObservationBuilder[np.ndarray] np_random : RandomState mu : np.ndarray
mean of appropriate size, defaults to 0
- sigmanp.ndarray
sigma of appropriate size, defaults to 1
Returns#
observation with Gaussian noise added