Systems Engineering for Autonomous Vehicles; Supervising AI using Large Language Models (SSuperLLM)

TL;DR

The paper investigates using Large Language Models to augment Systems Engineering for autonomous vehicles through requirements development, auditing, and supervisory control. It proposes a hierarchical, LLM-driven approach to craft and verify requirements, coupled with a supervisory AV control architecture that uses an offboard LLM behavior database and a CONTEXT TRANSLATION layer. A proof-of-concept in a simple bicycle-model AV with LQR control demonstrates that frequent LLM decision intervals (0.5 s) support safe, timely responses, while longer intervals (2 s) can fail to prevent unsafe events. The work highlights potential productivity gains and explains why safety-by-design and human oversight remain essential, pointing to further work needed in LLM maturity and explicit requirement specification. Overall, this work provides a concrete pathway toward integrating LLMs into SysEng for AVs and emphasizes the importance of verification, explainability, and SOTIF in real-world deployments.

Abstract

Generative Artificial Intelligence (GAI) and the idea to use hierarchical models has been around for some years now. GAI has proved to be an extremely useful tool for Autonomous Vehicles (AVs). AVs need to perform robustly in their environment. Thus the AV behavior and short-term trajectory planning needs to be: a) designed and architected using safeguarding and supervisory systems and b) verified using proper Systems Engineering (SysEng) Principles. Can AV Systems Engineering also use Large Language Models (LLM) to help Autonomous vehicles (AV) development? This reader-friendly paper advocates the use of LLMs in 1) requirements (Reqs) development and 2) Reqs verification and 3) provides a proof-of-concept of AV supervisory control. The latter uses a simulation environment of a simple planar (bicycle) vehicle dynamics model and a Linear Quadratic Regulator (LQR) control with an LLM Application Interface (API). The Open-Source simulation SW is available from the author accessible to the readers so that they can engage into the AV stack, LLM API and rules, SysEng and Reqs and fundamental vehicle dynamics and control.

Figures (10)

• Figure 1: The V-model of the systems engineering process 0. VnV feedbacks back to architecture and Reqs.
• Figure 2: Part of the answer of ChatGPT-3.5 in the question “What are the basic manoeuvres in driving a car?”
• Figure 3: LLM for functional decomposition and Reqs coverage assessment. The “Mismatched req” implies Reqs that have been identified as being different and/or missing from the human decomposed Reqs.
• Figure 4: The figure shows the Autonomous vehicle (AV) stack (abstraction from 11121314). EGO derives from the Greek word Εγω, that means myself. It refers to the AV itself as an entity in the AV navigation task. The three red arrow connectors between BEHAVIOR PREDICTION, EGO BEHAVIOR and PLANNING, depict that all three are depending on each other. Consider an AV leaving a zebra crossing while an occluded pedestrian (agent) appears in the scene. The pedestrian anticipates the AV to continue its path and would plan to pass behind the AV. If the AV comes suddenly to a halt (to avoid a probable collision), then it is possible that the pedestrian will stumble on the AV that was expected to be ahead (c.f. Explanation from Waymo 15).
• Figure 5: LLM aided AV navigation using a supervisory monitor. The whole architecture focuses on the EGO BEHAVIOR of Fig. \ref{['fig:Fig_3']} however, the same concept can be applied in the other modules. The three red blocks depict the ones where most of the LLM functionality resides.