Expanding the Classical V-Model for the Development of Complex Systems Incorporating AI

TL;DR

Classical V-model is increasingly inadequate for complex AI-enabled systems that operate in open-world environments. The authors propose an iterative data-based V-model that fuses ODD-driven design, synthetic and real data, and continuous integration, while keeping safety argumentation customizable. They analyze VVM, Waymo, and Tesla to illustrate the spectrum of approaches and formulate a generic macro-framework that generalizes across system granularities. The result is a transferable blueprint for development, verification, and validation of autonomous, AI-enabled systems, with emphasis on safety, data-driven improvement, and open-world coverage.

Abstract

Research in the field of automated vehicles, or more generally cognitive cyber-physical systems that operate in the real world, is leading to increasingly complex systems. Among other things, artificial intelligence enables an ever-increasing degree of autonomy. In this context, the V-model, which has served for decades as a process reference model of the system development lifecycle is reaching its limits. To the contrary, innovative processes and frameworks have been developed that take into account the characteristics of emerging autonomous systems. To bridge the gap and merge the different methodologies, we present an extension of the V-model for iterative data-based development processes that harmonizes and formalizes the existing methods towards a generic framework. The iterative approach allows for seamless integration of continuous system refinement. While the data-based approach constitutes the consideration of data-based development processes and formalizes the use of synthetic and real world data. In this way, formalizing the process of development, verification, validation, and continuous integration contributes to ensuring the safety of emerging complex systems that incorporate AI.

Figures (5)

• Figure 1: Visualization of the overall methodology of the VVM project VVMOverall as an extension of the classical V-model designed for automated driving application with regard to a scenario-based problem decomposition and an appropriate safety argumentation.
• Figure 2: Simplified representation of Waymo's safety determination lifecycle, inspired by favaro2023building. It illustrates the distinguished consideration of prospective and retrospective perspectives on the methodology and safety argumentation.
• Figure 3: Tesla's data engine karpathy_cvpr21, visualized in V-model structure, represents a fully data-driven methodology that is tailored to AI systems and strives for efficient and effective continuous improvement.
• Figure 4: Visualization of the classical V-model and the analyzed innovative development processes with regard to methodological generality and system compatibility. The remaining gap is marked by the proposed iterative data-based V-model.
• Figure 5: The iterative data-based V-model, which formalizes and merges the various existing methods. The initial loop starts with the definition of the ODD. The explicit formalization of the process from the real world to simulation and back and the data-based characteristic address the challenges of complex systems that embrace AI. The iterative approach, on the other hand, addresses the challenges of open world complexity and offers continuous system and confidence improvement in an intuitive way.