A modular system architecture for autonomous robots based on blackboard and publish-subscribe mechanisms

We present a system software architecture for mobile robots such as autonomous vehicles. The system achieves the goals of flexibility, maintainability, testability, and modifiability through a decoupled software architecture based on an asynchronous publish-subscribe mechanism and a blackboard object handling synchronized access to shared data. We report on two implementations using the proposed generic architecture and the POSIX real time API. The first implementation is for an autonomous vehicle using waypoint-based navigation, and the second implementation uses the same high-level modules but replaces the low-level hardware interfaces with a virtual reality simulation. Our experiments and an evaluation indicate that the architecture is suitable for a wide variety of control algorithms and supports the construction of testable, maintainable, and modifiable autonomous robot vehicles at low cost in terms of real-time performance.