APIs: common

common.registry

class common.registry.Registry(name: str = 'Registry', **kwargs)

A helper class for registering and managing modules

>>> # build registry
>>> REGISTRY_NAME = Registry(name='REGISTRY_NAME')

>>> # register function
>>> @REGISTRY_NAME.register
>>> def foo():
>>>     pass

>>> # register class
>>> @REGISTRY_NAME.register
>>> class bar():
>>>     pass

>>> # fetch for class construction within builder
>>> class_instance = REGISTRY_NAME[module_name](*args, **kwargs)

>>> # fetch for function call
>>> result = REGISTRY_NAME[module_name](*args, **kwargs)
__getitem__(name: str) → Any

Return self[key].

__init__(name: str = 'Registry', **kwargs)
Parameters:

name – name of the registry

__weakref__

list of weak references to the object (if defined)

register(module: Any) → Any

Register module (class/function/etc.) into this registry.

Typically used as decorator, thus return the input module itself.

Parameters:

module – Python object that needs to be registered.

Returns:

The registered module.

Return type:

Any

register_from_python_module(module: Any) → Self

Register all members from a Python module.

Parameters:

module – Python module to be processed.

Returns:

The yielded registry.

Return type:

Self

common.registry.build_object_within_registry_from_config(registry: Registry, config: Dict[str, Any] = {}, **kwargs) → Any

Builder function to build object within a registry from config.

Config should be in form of keyword arguments (dict-like). Support adding additional config items through kwargs.

NOTE: kwargs will not be deep-copied.

Parameters:

  • registry – registry to retrieve class to be constructed.

  • config –

    config function that provide the class name and the corresponding arguments, which should be arranged in the following format:

    type: TYPENAME
arg1: value1
arg2: value2
...

  • **kwargs – key-word arguments to be passed to the retrieved class function.

Returns:

The built object.

Return type:

Any

common.io

common.io.convert_TensorFP_to_numpy(tensor_proto: TensorFP) → ndarray

Convert tensor proto to numpy array.

Parameters:

tensor_proto – Tensor proto.

Returns:

Converted numpy array.

Return type:

np.ndarray

common.io.convert_list_to_tuple_within_dict(dictionary: Dict, exceptions: Tuple[str] = ()) → Dict

Recursively cast list to tuple within a dict.

Avoid modification issue (mutable / immutable). Support specified exception key.

Also deal with some type issues. e.g. in case of stable-baselines3, https://github.com/DLR-RM/stable-baselines3/blob/v2.2.1/stable_baselines3/common/off_policy_algorithm.py#L157

Parameters:

  • dictionary – Dictionary to be processed.

  • exceptions – Exceptional keys.

Returns:

Processed dictionary.

Return type:

Dict

common.io.convert_numpy_to_TensorFP(arr: ndarray) → TensorFP

Convert numpy array to tensor proto.

Parameters:

arr – Numpy array to be converted.

Returns:

Converted tensor proto.

Return type:

TensorFP

common.io.generate_random_string(n: int) → str

Generate a string with characters randomly chosen.

Parameters:

n – Desired length of string.

Returns:

Random string.

Return type:

str

common.io.load_and_override_configs(config_paths: List[str]) → Dict

Load and override a series of config files.

Parameters:

config_paths –

The base config and the overriding configs.

  • The first config will serve as base config.

  • The rest configs will override the base config, respectively.

Returns:

The loaded and overridden config.

Return type:

Dict

common.io.load_config_from_yaml(config_file: str) → Dict

Load config from yaml and handle exceptions.

Parameters:

config_file – Path to configuration file.

common.io.override_config(base_config: Dict, update_config: Dict, allow_new_key: bool = False) → Dict

Override the value config.

Parameters:

  • base_config – Base config to be processed.

  • update_config – Incremental config which contains key-value pair for overriding.

  • allow_new_key – Whether allow creation of new key.

Returns:

The overridden config.

Return type:

Dict

common.io.save_config_to_yaml(config: Dict, config_file: str)

Save config to YAML file.

Parameters:

  • config – Config object to be saved.

  • config_file – Path to YAML file.

common.gym_helper

common.gym_helper.scale_action(action: ndarray, action_space: Space) → ndarray

Scale action into range [-1, +1].

Parameters:

  • action – Action to be sclaled.

  • action_space – Reference action space.

Returns:

Scaled action.

Return type:

np.ndarray

common.geometry

common.geometry.normalize_angle(angle: float | ndarray) → float | ndarray

Normalize angle to [-pi, pi), compatible with Numpy vectorization.

Parameters:

angle – Angle to be normalized.

Returns:

Normalized angle.

common.geometry.transform_between_local_and_world(points: ndarray, body_state: ndarray, trans_dir: str) → ndarray

Transform points between the local (body) frame and the world frame.

Parameters:

  • points (np.ndarray) – Points to be transformed, shape=(N, 2).

  • body_state (np.ndarray) – Body state, format=<x, y, r>.

  • trans_dir (str) – Transform direction. ‘world_to_local’: from world frame to body frame. ‘local_to_world’: from body frame to world frame.

Returns:

Transformed points, shape=(N, 2).

Return type:

np.ndarray

common.geometry.transform_points(points: ndarray, transform_matrix: ndarray) → ndarray

Transform 2-D points with transform matrix.

Parameters:

  • points – Points to be transformed, shape=(N, 2).

  • transform_matrix – Transform matrix which lies in group SO(2), shape=(3, 3).

Returns:

Transformed points, shape=(N, 2).

Return type:

np.ndarray

common.geometry.transform_to_local_from_world(points: ndarray, body_state: ndarray) → ndarray

Transform points from the world frame to the local (body) frame.

Parameters:

  • points (np.ndarray) – Points in world frame, shape=(N, 2).

  • body_state (np.ndarray) – Body state, format=<x, y, r>.

Returns:

Points in body frame, shape=(N, 2).

Return type:

np.ndarray

common.geometry.transform_to_world_from_local(points: ndarray, body_state: ndarray) → ndarray

Transform points from the local (body) frame to the world frame.

Parameters:

  • points (np.ndarray) – Points in body frame, shape=(N, 2).

  • body_state (np.ndarray) – Body state, format=<x, y, r>.

Returns:

Points in world frame, shape=(N, 2).

Return type:

np.ndarray