Skip to content

Environment Configuration

The environment configuration contains a set of parameters that define how a bluebird_gymnasium environment is set up when instantiated. It is defined in the class bluebird_gymnasium.envs.EnvConfig.

Default Configuration

Each environment contains a default environment configuration which can be retrieved by calling the class method get_default_env_config.

The sample code snippet below fetches the default configuration for the SectorXPlusEnv environment.

from bluebird_gymnasium.envs import SectorXPlusEnv
env_config = SectorXPlusEnv.get_default_env_config()

Instantiate an Environment

bluebird_gymnasium environments are instantiated with an environment configuration. If unset at the environment instance creation, the default configuration is used.

The sample code snippet below demonstrates how to instantiate the SectorXPlusEnv environment.

from bluebird_gymnasium.envs import SectorXPlusEnv
env_config = SectorXPlusEnv.get_default_env_config()

# method 1: direct object creation interface
env_1 = SectorXPlusEnv(config=env_config)

# method 2: gymnasium `make` function interface
import gymnasium as gym
import bluebird_gymnasium.envs
env_2 = gym.make("SectorXPlusEnv-v0", config=env_config)

Save and Load Configurations from Disk

The configuration can be saved to and loaded from a disk as a JSON file.

The sample code snippet below saves a configuration to disk.

# assuming env_config was previously defined
import json
with open("env_config.json", "w") as fp:
    json.dump(vars(args), fp, indent=4)

The sample code snippet below loads a configuration from disk.

import json
from bluebird_gymnasium.envs import EnvConfig
with open("env_config.json", "r") as fp:
    env_config_dict = json.load(fp)
    env_config = EnvConfig(**env_config_dict)

Configuration Components

Each configuration comprises sub-configurations, primarily defined as dictionaries to configure various aspects of the environment.

Agent Focused Sub-Configurations

  • .action_config: defines the action space for an environment instance.
  • .reward_config: defines a set of reward functions and their respective weights used to compute the scalarized reward per aircraft per step.
  • .state_repr_config: defines the state encoding method to use in representing each aircraft state.
  • .view_config: defines the parameters that set up the gymnasium environment for either a single-agent (centralized) or multi-agent (decentralized).
  • .forward_fixes_config: defines the parameters used to configure an aircraft's route information based on the number of forward fixes. The config is exploited by the actions execution and aircraft state representation.
  • .airspace_config: defines the airspace parameters used by the underlying bluebird_dt digital twin.
  • .radar_config: defines the parameters used for visualizing the digital twin via a matplotlib-based or SVG plot.
  • .scenario_config: defines the parameters used to instantiate a scenario to run in the bluebird_dt digital twin.
  • .simulation_log_config: defines parameters used for specifying logging in the digital twin.

Other Non Sub-Configuration Parameters

Other parameters in the environment configuration include: - .scenario_duration: defines the total time (in seconds) to run a scenario within an episode. - .scenario_sec_per_step: defines the time duration (in seconds) to elapse per simulation step. Defaults to 6. Note, the number of RL steps within an episode is computed as: scenario_duration / scenario_sec_per_step. - .diagnostic_level: