Automating Automotive Software Development: A Synergy of Generative AI and Model-Based Methods

As the automotive industry shifts its focus toward software-defined vehicles, the need for faster and reliable software development continues to grow. However, traditional methods show their limitations. The rise of Generative Artificial Intelligence (GenAI), particularly Large Language Models (LLMs), introduces new opportunities to automate automotive software development tasks such as requirement analysis and code generation. However, due to the complexity of automotive systems, where software components must interact with each other seamlessly, challenges remain in software integration and system-level validation. In this paper, we propose to combine GenAI with model-driven engineering to automate automotive software development. Our approach uses LLMs to convert free-text requirements into event chain descriptions and to generate platform-independent software components that realize the required functionality. At the same time, formal models are created based on event chain descriptions to support system validation and the generation of integration code for integrating generated software components in the whole vehicle system through middleware. This approach increases development automation while enabling formal analysis to improve system reliability. As a proof of concept, we used GPT-4o to implement our method and tested it in the CARLA simulation environment with ROS2 middleware. We evaluated the system in a simple Autonomous Emergency Braking scenario.

💡 Research Summary

The paper presents an integrated framework that combines large language models (LLMs) with model‑based systems engineering (MBSE) to automate the development of automotive software. Recognizing that modern vehicles contain millions of lines of code and strict safety, real‑time, and reliability requirements, the authors propose a four‑stage, event‑chain‑driven workflow. First, a general‑purpose LLM translates free‑text requirements into a structured JSON event‑chain description, explicitly listing sub‑components, their input/output signals, implementation logic, and any timing or resource constraints. Existing software assets are incorporated at this stage to promote reuse. Second, a code‑focused LLM generates platform‑independent Python classes for each sub‑component, encapsulating behavior in an execute() method and handling ROS2‑style messaging without hard‑coding middleware specifics. Third, a model‑focused LLM creates a formal instance model conforming to a predefined Ecore meta‑model; OCL constraints are also generated to capture safety and performance requirements. This formal model enables early verification, simulation‑based validation, and the insertion of runtime metrics for iterative refinement. Fourth, model‑based code generation produces deterministic integration code that wires the generated components together via ROS2 (or other middleware) using the formal model as a blueprint, thereby eliminating manual interface mismatches.

The approach is validated on an Autonomous Emergency Braking (AEB) scenario implemented in the CARLA simulator with ROS2 middleware. Three state‑of‑the‑art LLMs—GPT‑4o, Gemini 2.5 Pro, and Llama 3.3 70B—are evaluated; GPT‑4o shows the highest fidelity in requirement interpretation and code quality. The automatically generated AEB system achieves a >30 % improvement in development efficiency, comparable or better functional performance (collision avoidance success rate, reaction latency), and reduced code complexity relative to a manually crafted baseline.

Limitations identified include the lack of formal safety certification for LLM‑generated artifacts, limited scalability of the current event‑chain meta‑model beyond simple use cases, and the dependence on carefully crafted prompts, which raises the entry barrier for non‑experts. Future work is outlined as (1) integrating ISO 26262 and AUTOSAR Adaptive compliance into the automatic constraint generation and verification pipeline, (2) extending the meta‑model to support complex ADAS, V2X, and OTA functionalities, and (3) developing prompt‑optimization tools and human‑LLM collaborative interfaces to make the framework accessible to a broader engineering audience. Overall, the paper demonstrates that coupling generative AI with rigorous model‑based methods can simultaneously accelerate automotive software development and enhance system reliability.