environments

The software includes single agent task simulation environments such as Atari, Mujoco, Classic Control, and Box2D under gym. It also includes multi-agent task simulation environments such as MPE and SISL under the open-source environment PettingZoo, StarCraft2, MAgent2, Google Football, etc. Each simulation environment contains a rich variety of task scenarios, as shown in the table below.

• single_agent_env.

• multi_agent_env.

• vector_envs.

• utils.

Make Environment

xuance.environment.make_envs(config: Namespace)

Creates and returns a set of environments based on the provided configuration.

This function supports single-agent, multi-agent, and vectorized environments and handles the initialization of the environment(s) based on the configuration settings. The function also manages distributed training setups and environment vectorization.

Parameters:

configNamespace

A configuration object containing the necessary settings to initialize the environment. The configuration should contain the following attributes: - env_name (str): The name of the environment to create. - env_seed (int): The seed value for environment initialization. - distributed_training (bool): Whether to use distributed training. - parallels (int): The number of parallel environments for vectorized setups. - vectorize (str): The type of vectorization to apply (e.g., ‘DummyVecEnv’, ‘SubprocVecEnv’, etc.).

returns:

List of environments based on the configuration settings.