Check if value is a numpy array, if so, convert it to a torch tensor.

Get the value of the specified property from the JSBSim simulation :param prop: Property :return : float

Flatten the first two dimensions of a tensor.

(north, east, up), unit: m

(v_north, v_east, v_up), unit: m/s

Set the values of the specified property :param prop: Property :param value: float

Args: time(float): second, max precision: 0.001s (recommend 0.01s/frame) state_in(list): [lon, lat, alt, roll, pitch,

Convert torch.Tensor to numpy.ndarray.

Get the values of the specified properties :param props: list of Properties : return: NamedTupl e with properties name and their val

Transform normalized data back into original distribution

Convert from Geodetic Coordinate System to NEU Coordinate System. Args: lon, lat, alt (float): target geodetic lontitude(°), latitude(°),

Pack seperated key-value dict into grouped np.ndarray

This function is used for buffer data operation. Specifically, it transposes a tensor from (episode_length, n_rollout_threads, *dim) to (n_rollout

Returns the available actions for agent_id.

Get shape from observation space. Args: obs_space: (gym.spaces or list) observation space Returns: obs_shape: (tuple) observat

Distance between two points.

Get unit by ID.

Get actions for rollout. Args: obs: (np.ndarray) input observation, shape is (n_threads, n_agents, dim) available_acti

Returns the ID of unit type in the given scenario.

Returns maximal shield for a given unit.

Evaluate the model.

Get AO & TA angles and relative distance between two agent. Args: ego_feature & enemy_feature (tuple): (north, east, down, vn, ve, vd)

Turn off grad for actor parameters.

Get value function predictions. Args: cent_obs: (np.ndarray) centralized input to the critic. rnn_states_critic: (np.n

Update the given property for all engines :param sim: simulation to use :param prop: property to update

Returns all agent observations in a list. NOTE: Agents should have access only to their local observations during decentralised execu

Process reward and inner variables. Args: new_reward (float) agent_id (str) render_items (tuple, optional

Compute actions using the given inputs. Args: obs: (np.ndarray) local agent inputs to the actor. rnn_states_actor: (np

Get the initialization method. Args: initialization_method: (str) initialization method Returns: initialization method: (torch

Returns all agent observations in a list. NOTE: Agents should have access only to their local observations during decentralised execut

Decay the actor and critic learning rates. Args: episode: (int) current training episode. episodes: (int) total number

Renders the environment. The set of supported modes varies per environment. (And some environments do not support rendering at all.)

Reset all the environments and return an array of observations, or a dict of observation arrays. If step_async is still doin

Turn on grad for actor parameters.

Decreases the learning rate linearly Args: optimizer: (torch.optim) optimizer epoch: (int) current epoch total_num_epochs:

Get action logprobs, entropy, and distributions for actor update. Args: obs: (np.ndarray / torch.Tensor) local agent inputs to the

Training data generator for actor that uses MLP network.

Get the activation function. Args: activation_func: (str) activation function Returns: activation function: (torch.nn) activat

Returns the size of the observation.

Init module. Args: module: (torch.nn) module weight_init: (torch.nn) weight init bias_init: (torch.nn) bias init g

Training data generator for actor that uses RNN network. This generator does not split the trajectories into chunks, and therefore ma

Training data generator for actor that uses RNN network. This generator splits the trajectories into chunks of length data_chunk_length,

Returns the sight range for an agent.

This function is used for buffer data operation. Specifically, it transposes a tensor from (episode_length, n_rollout_threads, num_agents, *dim) t

Whether a unit can move in a given direction.

Compute action log probability, distribution entropy, and action distribution. Args: x: (torch.Tensor) input to network.

Full restart. Closes the SC2 process and launches a new one.

Get actions and logprobs of actions for observations. Args: obs: (np.ndarray) observations of actor, shape is (batch_size, dim)

Get gradient norm.

Get mean rewards of the buffer

Returns the size of the vector containing the agents's movement-related features.

(roll, pitch, yaw), unit: rad

Returns height values of the grid surrounding the given unit.

Returns pathing values of the grid surrounding the given unit.

Convert individual actions to joint action. Args: orig_action: (int) joint action. Returns: actions: (list) in

Save a replay.