Toward Fully Autonomous Driving: AI, Challenges, Opportunities, and Needs

TL;DR

This paper analyzes the current state of automated driving (AD) and its AI-driven components, arguing that the modular service-oriented (SO) AD stack must evolve to achieve fully autonomous driving in open-world environments. It emphasizes situation awareness (Levels 1–3) and the benefits and risks of end-to-end learning and foundation models, proposing a conceptual SO-M-E2E architecture that combines modularity with data-driven orchestration via attention-based interfaces and external context sourcing. The authors discuss safety assurance, governance, and regulatory challenges, outlining a data-centric iterative development lifecycle and potential transferability strategies across different operational design domains. The work provides a forward-looking framework that balances interpretability, safety, and scalability, highlighting the need for contextual information, external data integration, and robust safeguarding to realize practical, trustworthy autonomous mobility.

Abstract

Automated driving (AD) is promising, but the transition to fully autonomous driving is, among other things, subject to the real, ever-changing open world and the resulting challenges. However, research in the field of AD demonstrates the ability of artificial intelligence (AI) to outperform classical approaches, handle higher complexities, and reach a new level of autonomy. At the same time, the use of AI raises further questions of safety and transferability. To identify the challenges and opportunities arising from AI concerning autonomous driving functionalities, we have analyzed the current state of AD, outlined limitations, and identified foreseeable technological possibilities. Thereby, various further challenges are examined in the context of prospective developments. In this way, this article reconsiders fully autonomous driving with respect to advancements in the field of AI and carves out the respective needs and resulting research questions.

Figures (7)

• Figure 1: High-level illustration of an exemplary modular service-oriented software architecture for AD on the sensor/actor level.
• Figure 2: Comparison of different driving situations with comparably perceived free space. In case (a), the context of the situation allows the free space to be used. In situation (b), however, caution is advised. The differences result from the context. This illustrates that not all free space is equal.
• Figure 3: Endsley's model of situation awareness in dynamic decision making according to endsley1995toward.
• Figure 4: The spectrum of scenarios in autonomous driving in the real world is described as an open long-tail distribution. In addition to scenarios that occur very frequently, there are a large number of rare scenarios and the challenge of novel scenarios that arise due to the characteristics of the open world.
• Figure 5: General overview of architectural AD stack approaches aligned with the SA model endsley1995toward.