PolyFlows: Modular Test Framework Design for Autonomous Vehicles

Validation and verification of autonomous vehicles is a vexing research challenge. PolyVerif, an open-source validation and verification framework, is designed to accelerate research in this area. PolyVerif presents a comprehensive open-source validation and verification stack, including an autonomous vehicle engine, symbolic test generation, simulation at different fidelity levels, a digital twin creation environment, and a process for transitioning from simulation tests to physical tests. One of the key challenges for the validation and verification of autonomous vehicles is to construct a modular design for the test flows. Modular validation is critical to building an inductive argument for system correctness. Today, these techniques are not easily available to validation researchers or engineers due to the lack of visibility into the internals of an autonomous vehicle stack and the lack of standard definitions for these interface points. The realm of open-source allows for both visibility and experimentation to develop these very important interfaces. This paper introduces PolyFlows, the modular "Design for Test" framework in PolyVerif, focused on the main pieces of the complex autonomous vehicle stack: Detection, Location Services, Control, and Path Planning. PolyFlows creates open-source modular validation structures that can be used to check the precision of the intricate autonomous vehicle stack. The paper also presents the use of this framework in the Jacksonville Transportation Authority Ultimate Urban Circulator project, which is contributed to the open-source PolyVerif repository.