[Paper] Automotive Engineering-Centric Agentic AI Framework

We just been working to elaborate the role of AI agents across workflow, from offline learning to online tuning, reasoning, and control actions (e.g., parameter updates, simulation toolchain execution). We also propose a control-inspired framework, structuring workflows as feedback systems guided by workflow stability, using error signals such as performance gap, constraint violation, and simulation cost (also inspried from Lyapunov-like energy function :).

Several concrete industrial use cases (with colleagues from Siemens) are also presented:

1. multimodal engineering data learning
2. suspension design
3. reinforcement learning for control tuning
4. surrogate-assisted aerodynamic optimization
5. MBSE evolution and updates

This is a nice joint effort from multiple Siemens Simcenter Engineering teams across different locations: Yerlan Akhmetov, Piero Brigida, Zhihao Liu, Gurudevan Devarajan, Ajinkya Bhave, Kai Liu.

We will also present it at the JSAE Annual Congress in Japan this May.

Hope you enjoy the reading :) . This is a practical perspective on the real impact of Agentic AI in engineering.

Link: https://arxiv.org/pdf/2604.07784