cambrian.envs.maze_env

Defines the MjCambrianMazeEnv class.

Classes

MjCambrianMapEntity	Enum representing different states in a grid.
MjCambrianMazeConfig	Defines a map config. Used for type hinting.
MjCambrianMazeEnvConfig	mazes (Dict[str, MjCambrianMazeEnvConfig]): The configs for the mazes. Each
MjCambrianMazeEnv	A MjCambrianEnv defines a gymnasium environment that's based off mujoco.
MjCambrianMaze	The maze class. Generates a maze from a given map and provides utility
MjCambrianMazeStore	This is a simple class to store a collection of mazes.

Module Contents

class MjCambrianMapEntity(*args, **kwds)

Bases: enum.Enum

Enum representing different states in a grid.

Variables:

• RESET (str) – Initial reset position of the agent. Can include agent IDs in the format “R:<agent id>”.

• WALL (str) – Represents a wall in the grid. Can include texture IDs in the format “1:<texture id>”.

• EMPTY (str) – Represents an empty space in the grid.

static parse(value)

Parse a value to handle special formats like “1:<texture id>”.

Parameters:

value (str) – The value to parse.

Returns:

Tuple[Self, str] –

The parsed entity and the texture id if

applicable.

class MjCambrianMazeConfig

Bases: hydra_config.HydraContainerConfig

Defines a map config. Used for type hinting.

Variables:

• xml (MjCambrianXML) – The xml for the maze. This is the xml that will be used to create the maze.

• map (str) – The map to use for the maze. It’s a 2D array where each element is a string and corresponds to a “pixel” in the map. See maze.py for info on what different strings mean. This is actually a List[List[str]], but we keep it as a string for readability when dumping the config to a file. Will convert to list when creating the maze.

• scale (float) – The maze scaling for the continuous coordinates in the MuJoCo simulation.

• height (float) – The height of the walls in the MuJoCo simulation.

• hflip (bool) – Whether to flip the maze or not. If True, the maze will be flipped along the x-axis.

• vflip (bool) – Whether to flip the maze or not. If True, the maze will be flipped along the y-axis.

• rotation (float) – The rotation of the maze in degrees. The rotation is applied after the flip.

• wall_texture_map (Dict[str, List[str]]) – The mapping from texture id to texture names. Textures in the list are chosen at random. If the list is of length 1, only one texture will be used. A length >= 1 is required. The keyword “default” is required for walls denoted simply as 1 or W. Other walls are specified as 1/W:<texture id>.

• agent_id_map (Dict[str, List[str]]) – The mapping from agent id to agent names. Agents in the list are chosen at random. If the list is of length 1, only one agent will be used. A length >= 1 is required for each agent name. Effectively, this means you can set a reset position as R:<agent id> in the map and this map is used to assign to a group of agents. For instance, R:O in the map and including O: [agent1, agent2] in the agent_id_map will assign the reset position to either agent1 or agent2 at random. “default” is required for agents denoted simply as R.

• enabled (bool) – Whether the maze is enabled or not.

class MjCambrianMazeEnvConfig

Bases: cambrian.envs.env.MjCambrianEnvConfig

mazes (Dict[str, MjCambrianMazeEnvConfig]): The configs for the mazes. Each

maze will be loaded into the scene and the agent will be placed in a maze at each reset.

maze_selection_fn (MjCambrianMazeSelectionFn): The function to use to select

the maze. The function will be called at each reset to select the maze to use. See MjCambrianMazeSelectionFn and maze.py for more info.

class MjCambrianMazeEnv(config, **kwargs)

Bases: cambrian.envs.env.MjCambrianEnv

A MjCambrianEnv defines a gymnasium environment that’s based off mujoco.

NOTES: - This is an overridden version of the MujocoEnv class. The two main differences is that we allow for /reset multiple agents and use our own custom renderer. It also reduces the need to create temporary xml files which MujocoEnv had to load. It’s essentially a copy of MujocoEnv with the two aforementioned major changes.

Parameters:

• config (MjCambrianEnvConfig) – The config object.

• name (Optional[str]) – The name of the environment. This is added as an overlay to the renderer.

generate_xml()

Generates the xml for the environment.

reset(*, seed=None, options=None)

Reset the environment.

Will reset all underlying components (the maze, the agents, etc.). The simulation will then be stepped once to ensure that the observations are up-to-date.

Returns:

Tuple[ObsType, InfoType] –

The observations for each

agent and the info dict for each agent.

property maze: MjCambrianMaze

Returns the current maze.

property maze_store: MjCambrianMazeStore

Returns the maze store.

class MjCambrianMaze(config, name)

The maze class. Generates a maze from a given map and provides utility functions for working with the maze.

reset(spec, *, reset_occupied=True)

Resets the maze. Will reset the wall textures and reset the occupied locations, if desired.

compute_optimal_path(start, target, *, obstacles=[])

Computes the optimal path from the start position to the target.

Uses a BFS to find the shortest path.

Keyword Arguments:

obstacles (List[Tuple[int, int]]) – The obstacles in the maze. Each obstacle is a tuple of (row, col). Defaults to []. Avoids these positions when computing the path.

generate_reset_pos(agent, *, add_as_occupied=True)

Generates a random reset position for an agent.

Keyword Arguments:

add_as_occupied (bool) – Whether to add the chosen location to the occupied locations. Defaults to True.

Returns:

np.ndarray – The chosen position. Is of size (2,).

property config: MjCambrianMazeEnvConfig

Returns the config.

property name: str

Returns the name.

property map: numpy.ndarray

Returns the map.

property map_length_scaled: float

Returns the map length scaled.

property map_width_scaled: float

Returns the map width scaled.

property max_dim: float

Returns the max dimension.

property min_dim: float

Returns the min dimension.

property ratio: float

Returns the ratio of the length over width.

property x_map_center: float

Returns the x map center.

property y_map_center: float

Returns the y map center.

property lookat: numpy.ndarray

Returns a point which aids in placement of a camera to visualize this maze.

property reset_locations: List[numpy.ndarray]

Returns the reset locations.

class MjCambrianMazeStore(maze_configs, maze_selection_fn)

This is a simple class to store a collection of mazes.

generate_xml()

Generates the xml for the current maze.

reset(spec)

Resets all mazes.

property current_maze: MjCambrianMaze

Returns the current maze.

property maze_list: List[MjCambrianMaze]

Returns the list of mazes.

select_maze(env)

This should be called by the environment to select a maze.

select_maze_random(_)

Selects a maze at random.

select_maze_schedule(env, *, schedule='linear', total_timesteps, n_envs, lam_0=-2.0, lam_n=2.0)

Selects a maze based on a schedule. The scheduled selections are based on the order of the mazes in the list.

Keyword Arguments:

• schedule (Optional[str]) – The schedule to use. One of “linear”, “exponential”, or “logistic”. Defaults to “linear”.

• total_timesteps (int) – The total number of timesteps in the training schedule. Unused if schedule is None. Required otherwise.

• n_envs (int) – The number of environments. Unused if schedule is None. Required otherwise.

• lam_0 (Optional[float]) – The lambda value at the start of the schedule. Unused if schedule is None.

• lam_n (Optional[float]) – The lambda value at the end of the schedule. Unused if schedule is None.

select_maze_cycle(env)

Selects a maze based on a cycle.