Skip to content

Autoware concepts#

The concept of Autoware revolves around providing an open and flexible platform to accelerate the development and deployment of autonomous driving systems. This page explains its two key ideas: the microautonomy architecture and the Core and Universe repository model.

Microautonomy architecture#

Microautonomy architecture is the way Autoware breaks down "autonomous driving" into many small, replaceable capabilities rather than one monolithic stack. Each capability (for example, object detection, behavior planning, or lane-level routing) is a module with clear inputs and outputs, so systems can be composed like building blocks for different vehicles and use cases.

What is microautonomy architecture?

In short, it is a component-based autonomy design where driving behaviors are built by composing many small autonomy modules instead of relying on a single, fixed pipeline. This makes it easy to mix, match, and upgrade parts without rewriting the whole system.

Autoware's modules are connected through well-defined interfaces, which allows you to swap or extend individual components while keeping the rest of the system intact. For example, you can replace the default object detection with a custom neural network specialized in construction cones, and the downstream tracking, planning, and control modules still work as before.

  1. Start with the default perception pipeline.
  2. Plug in a dedicated detector for special targets, for example cones or forklifts.
  3. Keep the same planner and controller. They see "objects", regardless of how they were detected.

At a high level, these interfaces come in two flavors:

  • External AD APIs

    Expose Autoware's capabilities to the outside world, for example fleet management, cloud services, or infotainment.

Why this matters for developers and partners#

  • Reuse

    You can reuse core Autoware modules and only customize what is unique to your product.

  • Evolve step by step

    You can evolve your stack incrementally and swap one module at a time.

  • Collaborate

    Partners can collaborate around shared, stable interfaces and contribute components that plug into a common ecosystem.

Core and Universe repository model#

Autoware's software ecosystem is organized in layers: Autoware Core, Autoware Universe, and the community packages in the Autoware Index. Together, they balance quality assurance with community-driven innovation.

Dependencies point inward: Universe depends on Core, and Core never depends on Universe.

Quality-assured base

Autoware Core#

Autoware Core contains the foundational packages maintained by the Autoware Foundation (AWF). These packages follow strict development standards: unit tests, a coverage gate, a stricter static analysis profile, and CI on every supported ROS distribution. Core serves as the stable platform that users can rely on for building autonomous driving systems.

  • A vetted, reliable foundation
  • Consistent APIs and behavior
  • Maintained, tested, and versioned releases

Community innovation

Autoware Universe#

Autoware Universe is a broader collection of open-source packages contributed by individuals, companies, and research groups. These packages are owned and maintained by their original authors, who set their own quality and development practices.

  • Rapid prototyping and experimentation
  • A place for sharing specialized modules
  • An entry point for contributions from the global community

Contributions can take two forms:

  • Merged directly into the Universe repository hosted by the AWF.
  • Hosted in their own repositories and listed in the Autoware Index.

Universe acts as a sandbox for experimentation, allowing new ideas, algorithms, and hardware adaptations to be shared quickly.

From Universe to Core#

Promising packages from Autoware Universe can be adopted into Autoware Core when they demonstrate sufficient maturity, stability, and usefulness. This creates a natural pipeline:

  • Innovation in Universe
  • Standardization to the Core requirements
  • Production use from Core

Learn more#