This repo provides tools for creating a dataset from the Carla autonomous driving simulator. It provides a docker compose file for running both a carla server and a jupyter server for connecting the carla client. It also provides a data collector interface and torch dataset.
A jupyter notebook for data capture and data loading in PyTorch are provided.
The docker containers will run the carla server and client for interfacing with the carla simulator. Simply run docker compose build && docker compose up to run the containers. The client container will expose a port (specified in the docker-compose.yaml file, 8888 by default) to connect to the jupyer server using http://localhost:8888.
from carla_data_collector import CarlaDataCollector
# hostname is the hostname of the carla server, either localhost or the name of the docker container
# data_dir is the directory to store the data
data = CarlaDataCollector(hostname='carla_server', data_dir='/data')
data.add_ego_vehicle()
# sensor data will be output to /data_dir/label
data.add_rgb_camera(label='rgb')
data.add_depth_camera(label='depth')
data.set_ego_autopilot(True)
# will capture num_ticks datapoints from carla
data.start(num_ticks=1000)
data.stop()from carla_dataset import CarlaDataset
dataset = CarlaDataset('/data', transform=ToTensor())
imgs = next(iter(dataset))
rgb = imgs['rgb']
depth = imgs['depth']Currently, only image sensors (rgb, depth) are supported. The goal is to support all sensor types in the data collector and data loader with their respective formats