Novel AD/ADAS Virtual Validation with Integration of LLM and Digital Twin Technologies

Prof. Dr. Reza Rezaei･Dr. Alexander Bradler･Dr. Christian Lang･Yassine Neifer･Carsten Simon･Thorsten Scheibe (IAV)

This paper showcases a novel AI-driven method that understands natural language requirements and scenario specifications and videos to create high-fidelity simulations, thereby automating creation and modification of scenarios for enhanced AD/ADAS virtual validation. It comprises three key components: generating simulation models from textual inputs using an adapted large-language model (LLM); extracting scenarios from real test videos using visual LLM, including detection of corner cases like glare and camera soiling; and enhancing criticality with multi-agent approach. The results will be presented for above-mentioned use-cases, covering the spectrum from ADAS to full AD, including co-simulation of different tools based on JAMBE model connection. An assessment of the approach for efficiency enhancement along with an outlook will be provided.

Building trust in collaborative simulations

Henri Sohier･Romain Barbedienne (IRT SystemX)･Julien Silande (ESI)

Collaborative simulations often suffer from organizational silos, hindering trust due to challenges in information exchange about both the simulation and the system it represents. Capturing, controlling, and communicating critical data throughout the development cycle -from specification to validation- requires structured solutions. This paper proposes clear requirements and practical tools to address these challenges. Factors such as the decision-making context, barriers between stakeholders, and their varying levels of technical expertise are carefully considered. Our approach fosters confidence among stakeholders, with a particular focus on the automotive industry, where effective collaboration is crucial to managing system complexity.

Simulation verification and validation: the challenge we all face

Henri Sohier･Romain Barbedienne (IRT SystemX)･Julien Silande (ESI)

Simulation Verification and Validation (V&V) spans a wide range of subtopics and approaches, with nuanced distinctions in system engineering, software engineering, and simulation engineering. Recognized standards, such as those by NASA and NAFEMS, encapsulate various existing practices. V&V is critical to building trust, ensuring credibility, and maintaining quality in simulations and simulation architectures. This article examines methodologies and tools designed to optimize current best practices while fostering effective communication about V&V processes, addressing key challenges in achieving trustworthy and integrated simulation frameworks.

MBD推進向けたシステムズエンジニアリングの取り組み

平山　友貴・町田　拓也・湯村　洋典（スバル）

CASE対応等で自動車開発が急速に高度・複雑化しており、開発期間短縮やコスト削減が課題となっている。その対策の一つとして、モデルベース開発（MBD）による試作・評価の期間・費用低減が期待される。自動車の様々な領域でMBDが進展しているが、MBDプロセス構築には未だ課題が多い。要因の一つとして、開発のデジタル化とプロセス変更を同時に行う事の難しさがある。そこで我々は要求定義～モデル構築～評価のプロセスを既存の試作車の分担・手番と不変とし、“純粋な試作車のモデルへの置換え”に専念する事でMBDの定着に取り組んでいる。本取り組みでは、デジタル化した要求定義から検証までのプロセスを、ISO/IEC/IEEE 15288のシステムライフサイクルプロセスのフレームワークを使って検証した。既存の要求定義～試作車製作プロセスにおいてMBSE・MBDへ適用することで、意思決定領域を拡大した。