flatland.envs.rail_generators module

Rail generators: infrastructure manager (IM) / Infrastrukturbetreiber (ISB).

class flatland.envs.rail_generators.EmptyRailGen(*args, **kwargs)

Bases: RailGen

Returns a generator which returns an empty rail mail with no agents. Primarily used by the editor

generate(width: int, height: int, num_agents: int, num_resets: int = 0, np_random: RandomState | None = None) → Callable[[int, int, int, int, RandomState], Tuple[RailGridTransitionMap, Dict | None]]

class flatland.envs.rail_generators.FileRailFromGridGen(filename: Path)

Bases: RailGen

static from_dict(dict)

generate(width: int, height: int, num_agents: int, num_resets: int = 0, np_random: RandomState | None = None) → Tuple[RailGridTransitionMap, Dict | None]

static save(filename: Path, prod: Tuple[RailGridTransitionMap, Dict | None])

static to_dict(prod: Tuple[RailGridTransitionMap, Dict | None])

static wrap(rail_generator: Callable[[int, int, int, int, RandomState], Tuple[RailGridTransitionMap, Dict | None]], rail_pkl: Path) → Callable[[int, int, int, int, RandomState], Tuple[RailGridTransitionMap, Dict | None]]

class flatland.envs.rail_generators.RailFromGridGen(rail_map, optionals=None)

Bases: RailGen

generate(width: int, height: int, num_agents: int, num_resets: int = 0, np_random: RandomState | None = None) → Tuple[RailGridTransitionMap, Dict | None]

class flatland.envs.rail_generators.RailGen(*args, **kwargs)

Bases: object

Base class for RailGen(erator) replacement

WIP to replace bare generators with classes / objects without unnamed local variables which prevent pickling.

generate(width: int, height: int, num_agents: int, num_resets: int = 0, np_random: RandomState | None = None) → Tuple[RailGridTransitionMap, Dict | None]

flatland.envs.rail_generators.RailGeneratorProduct

A rail generator returns a RailGenerator Product, which is just a RailGridTransitionMap followed by an (optional) dict/

alias of Tuple[RailGridTransitionMap, Dict | None]

class flatland.envs.rail_generators.SparseRailGen(max_num_cities: int = 2, grid_mode: bool = False, max_rails_between_cities: int = 2, max_rail_pairs_in_city: int = 2, seed: int | None = None, p_level_free: float = 0)

Bases: RailGen

static argsort(seq)

Same as Numpy sort but for lists Parameters ———- seq: List

list that we would like to sort from smallest to largest

Returns

Returns the sorted list

static fix_transitions(grid_transition_map: GridTransitionMap, rcPos: List[Tuple[int, int]], direction: Tuple[int, int] = -1, random_generator: RandomState | None = None)

Fixes broken transitions

generate(width: int, height: int, num_agents: int, num_resets: int = 0, np_random: RandomState | None = None) → Tuple[RailGridTransitionMap, Dict | None]

Parameters

width: int

Width of the environment

height: int

Height of the environment

num_agents:

Number of agents to be placed within the environment

num_resets: int

Count for how often the environment has been reset

np_random: RandomState

Returns

Returns the grid_map –> The railway infrastructure Hints: agents_hints’: {

‘num_agents’: how many agents have starting and end spots ‘agent_start_targets_cities’: touples of agent start and target cities ‘train_stations’: locations of train stations for start and targets ‘city_orientations’ : orientation of cities

get_closest_neighbour_for_direction(closest_neighbours, out_direction)

Given a list of clostest neighbours in each direction this returns the city index of the neighbor in a given direction. Direction is a 90 degree cone facing the desired directiont. Example:

North: The closes neighbour in the North direction is within the cone spanned by a line going North-West and North-East

Parameters

closest_neighbours: List

List of length 4 containing the index of closes neighbour in the corresponfing direction: [North-Neighbour, East-Neighbour, South-Neighbour, West-Neighbour]

out_direction: int

Direction we want to get city index from North: 0, East: 1, South: 2, West: 3

Returns

Returns the index of the closest neighbour in the desired direction. If none was present the neighbor clockwise or counter clockwise is returned

flatland.envs.rail_generators.empty_rail_generator() → Callable[[int, int, int, int, RandomState], Tuple[RailGridTransitionMap, Dict | None]]

flatland.envs.rail_generators.rail_from_file(filename: str | Path | None = None, load_from_package=None, env_dict=None) → Callable[[int, int, int, int, RandomState], Tuple[RailGridTransitionMap, Dict | None]]

Utility to load rail generator from persisted env - uses env_dict if populated, otherwise tries to load from file / package. Beware: it only allows to re-generate the same rail again. A call to reset with a new seed will not generate a new rail. TODO flatland-association/flatland-rl#242 implement registry for all generators (rail, line, timetable, malfunction) similar to ray.tune.registry

Parameters

filenameUnion[str, Path]

pickle file with persisted env, defaults to None.

load_from_packagestr

package, defaults to None.

env_dict: dict

env_dict, defaults to None.

Returns

function

Generator function that always returns a RailGridTransitionMap object with the matrix of correct 16-bit bitmaps for each rail_spec_of_cell.

flatland.envs.rail_generators.rail_from_grid_transition_map(rail_map, optionals=None) → Callable[[int, int, int, int, RandomState], Tuple[RailGridTransitionMap, Dict | None]]

flatland.envs.rail_generators.sparse_rail_generator(*args: object, **kwargs: object) → Callable[[int, int, int, int, RandomState], Tuple[RailGridTransitionMap, Dict | None]]