Skip to content

Overview#

What Open AD Kit is, why it exists, and how it relates to Autoware — the autonomous driving stack it packages for cloud-native deployment.

What is Open AD Kit#

Open AD Kit packages the Autoware autonomous driving stack as a set of focused, independently deployable container images. Rather than shipping one monolithic image, it splits the stack along the AD pipeline — sensing, perception, localization, mapping, planning, control, API, simulation, visualization, and CARLA bridge services — so you can run only what a given workload needs.

Autoware provides the autonomy stack; Open AD Kit makes it deployable. It packages upstream software into composable container images, defines deployment configurations, integrates with target platforms and vehicle systems, and maintains the build, test, and release tooling needed to run consistently from simulation through in-vehicle deployment.

SOAFEE Blueprint

The first SOAFEE blueprint for the software-defined vehicle, co-developed with the eSync Alliance. Learn more on the Platforms page.

Why Open AD Kit#

Modular Components#

Independent images for each stage of the AD pipeline. Deploy only what you need.

Mixed Criticality#

Separate workloads by criticality assumption across safety-qualified and standard hardware.

Cloud Native#

Scale from simulation to the edge with Docker Compose, Docker Bake, and platform integrations.

Connected and Continuous#

CI/CD with GitHub Actions, optimized build caching, and containerized testing.

How It Works#

Open AD Kit runs Autoware as a pipeline of containerized components. Each container handles one stage of autonomous driving, and the stages communicate over ROS 2 DDS on the host network:

Single-host deployments bind CycloneDDS to loopback by default. Set CYCLONEDDS_NETWORK_INTERFACE to an exact LAN or VPN interface name only when cross-host DDS is required. ROS_DOMAIN_ID separates domains but does not provide authentication or encryption.

  1. Sensing captures and preprocesses raw sensor data (LiDAR, camera, IMU).
  2. Perception detects and tracks objects, traffic lights, and drivable space.
  3. Mapping serves high-definition map data that the rest of the stack consumes.
  4. Localization determines the vehicle's exact position on the map.
  5. Planning computes a safe, feasible trajectory to the goal.
  6. Control converts that trajectory into throttle, brake, and steering commands.
  7. Vehicle System bridges those commands to the actual vehicle or simulator.

A deployment combines the shared base services with task-specific overlays. Planning and Scenario Simulation add a dummy simulator on top of the base; Logging Simulation adds sensing, perception, and localization for recorded sensor data. For the full picture, see Components and Deployment.