Skip to content

Latest commit

 

History

History
49 lines (38 loc) · 2.28 KB

README.md

File metadata and controls

49 lines (38 loc) · 2.28 KB
Raw

Dockerized Components

Build all Docker Images

Use following command to build all Docker images you'll need for launching the CARLA simulation and the driving components.

docker-compose -f carla-sim-build.yml build
docker-compose -f driving-components-build.yml build

Note: Splitting up the components into CARLA simulation and driving was required to speed up the server-side GitHub workflows because building the Docker images from scratch on each commit would imply downloading ~15 GB of pre-built Docker images which is not feasible

Run the Integration Tests

Testing the driving components against the CARLA simulator might sometimes be a bit tricky. Therefore, the integration tests are mocking CARLA with pre-recorded ROS bag files. This might be especially useful for developing / testing on machines without NVIDIA GPU.

For preparing the integration test assets, run following command:

./prepare-integration-test-assets.sh

Run following command to execute the integration tests:

docker-compose -f integration-tests-compose.yml up --abort-on-container-exit

CARLA Simulation Components

The CARLA simulation involves following components:

  • CARLA simulator: runs the basic CARLA simulation environment including world rendering, etc.
  • CARLA ROS bridge: interfaces for information exchange between CARLA and ROS
    • Scenario Runner: launches and monitors well-defined scenarios
    • CARLA RVIZ: spectates the remote-controlled car and its sensors / actuators

Driving Components (ROS)

The components for driving are the following:

  • Global Planner: serves navigation tasks based on map data Details
  • Perception: preprocesses sensor data into high-level driving information Details
    • Traffic Light Detection: evaluates traffic lights captured by cameras
    • Object Detection: tracks stationary / moving objects like other cars / pedestrians and predicts their future movement
  • Local Planner: serves local decision-making tasks and amplifies the route based on sensor data Details
    • Vehicle Controller: transforms the planned trajectory into actionable remote-control signals