gym-racer

OpenAI gym environment of a racing car.

Install

The repo is uploaded as a package on pip, but it is not updated that often.

Clone the repo and install it locally

git clone https://github.com/Pitrified/gym-racer.git
cd gym-racer
pip install -e .

Environment

Action space

The action space is a combination of two discrete action spaces:

1) accelerate:  NOOP[0], UP[1], DOWN[2]
2) steer:  NOOP[0], LEFT[1], RIGHT[2]

Reward

The reward is mainly related to the direction along the street, then multiplied by the speed of the car squared.

Observation space

There are two types of observation possible:

• lidar with shape (n): a lidar with n rays, each element indicates the first sensor out of the road
• diamond with shape (n, n): a square of sensors, with a corner on the car

The observation type can be set by passing it in the call to gym.make, and the default parameters can be overridden by passing a sensor_array_params dict:

sat = "lidar"       # default
#  sat = "diamond"
racer_env = gym.make(
    "racer-v0",
    sensor_array_type=sat,
    sensor_array_params=sensor_array_params,
)

The default parameters for the lidar are:

sensor_array_params["ray_num"] = 7               # number of rays per side
sensor_array_params["ray_step"] = 15             # distance between sensors along a ray
sensor_array_params["ray_sensors_per_ray"] = 20  # number of sensors along a ray
sensor_array_params["ray_max_angle"] = 70        # angle to sweep left/right

The default parameters for the diamond are:

sensor_array_params["viewfield_size"] = 20  # number of rows/columns in the sensor
sensor_array_params["viewfield_step"] = 10  # spacing between the dots

Render modes

There are two types of render mode available, the human mode initializes pygame and renders what the car is doing to the screen, while in console mode only the bare minimum of the pygame environment is loaded (to use spritecollide). An environment in console mode cannot be rendered as human.

The render mode can be set by passing it in the call to gym.make:

mode = "human"       # default
#  mode = "console"
racer_env = gym.make(
    "racer-v0",
    render_mode=mode,
)

Info

Info is a dict with some car details:

keys = ["car_pos_x", "car_pos_y", "car_dir", "car_speed"]