Configuring Environments¶

The deepracer-v0 environment can be customized by specifying configurations in gym.make():

env = gym.make(
    "deepracer-v0",
    agent_config: dict=agent_config,    # action space + sensors
    track_config: dict=track_config,    # track, bots, obstacles
    reward_function=reward_function,    # custom reward function
    cpus: int=3, memory: str="6g",      # allocate sim resources
    cache: bool=True,                   # keep sim warm on close
)

Most users should only need to specify reward_function, agent_config, and track_config, each of which is described below.

Reward Function¶

The reward function accepts the AWS DeepRacer input parameters and returns a numeric reward. The same parameters are exposed in info["reward_params"]. You can use the packaged default reward function, or define a custom one:

Example

Complete lapKeep center (default)

def reward_function(params):
    """Example of rewarding the agent for lap completion"""

    if params["progress"] >= 100:
        reward = 1
    else:
        reward = 0

    return float(reward)

def reward_function(params):
    """Example of rewarding the agent to follow center line"""

    # Read input parameters
    track_width = params["track_width"]
    distance_from_center = params["distance_from_center"]

    # Calculate 3 markers that are at varying distances away from the center line
    marker_1 = 0.1 * track_width
    marker_2 = 0.25 * track_width
    marker_3 = 0.5 * track_width

    # Give higher reward if the car is closer to center line and vice versa
    if distance_from_center <= marker_1:
        reward = 1.0
    elif distance_from_center <= marker_2:
        reward = 0.5
    elif distance_from_center <= marker_3:
        reward = 0.1
    else:
        reward = 1e-3  # likely crashed/ close to off track

    return float(reward)

Tip

For more examples and design ideas, see the AWS DeepRacer catalog of reward function examples.

Agent Config¶

The agent_config defines the action and observation spaces. The settings of relevance include:

Parameter	Description
`action_space_type`	Can be `discrete` or `continuous`.
`action_space`	Defines the action space in terms of `speed` and `steering_angle`. See examples below.
`sensor`	Can be `FRONT_FACING_CAMERA` (a 160×120 colored image), `STEREO_CAMERAS` (two 160×120 greyscale images), and/or `LIDAR` (64 radial readings). See the AWS DeepRacer sensors page for details.

Example

ContinuousDiscrete (default)

Continuous actions with LiDAR + front-facing camera.

agent_config = {
    "action_space": {
        "steering_angle": {"high": 30, "low": -30},
        "speed":          {"high":  2, "low":   1},
    },
    "action_space_type": "continuous",
    "sensor": ["FRONT_FACING_CAMERA", "LIDAR"],
}

Discrete actions with LiDAR + stereo camera.

agent_config = {
    "action_space": [
        {"steering_angle":  30, "speed": 0.6},
        {"steering_angle":  15, "speed": 0.6},
        {"steering_angle":   0, "speed": 0.6},
        {"steering_angle": -15, "speed": 0.6},
        {"steering_angle": -30, "speed": 0.6},
    ],
    "action_space_type": "discrete",
    "sensor": ["STEREO_CAMERAS", "LIDAR"],
}

Based on the specification in agent_config, the action and observation spaces are defined using the Gymnasium API, as described in the Gymnasium Interface section of the user guide.

Warning

Two camera sensors cannot be selected at once, and LiDAR cannot be selected alone.

Track Config¶

The track_config defines the environment (the track, #obstacles, #bot cars, etc). For example:

track_config = {
    "WORLD_NAME":           "reInvent2019_wide",
    "NUMBER_OF_OBSTACLES":  "0",
    "NUMBER_OF_BOT_CARS":   "0",
    # ... and more.
}

Info

All track_config parameters are string-valued because the simulator service reads them as environment variables.

The WORLD_NAME parameter specifies the track, selected from the Track Catalog.

Please see the Track Configuration reference and/or the DeepRacer-for-cloud documentation for descriptions of available track_config parameters.