Citation :

Harihara sudhan, Sanjaykumar, 2025. "Autonomous Engineering Agents: Toward Self-Driving CAD and Infrastructure Design with AI in the Engineering Loop" ESP International Journal of Emerging Multidisciplinary Research [ESP-IJEMR]  Volume 1, Issue 1: 64-73.

Abstract :

On the one side, engineering design is taking on a new dimension as artificial intelligence (AI) and self-governing software agents are beginning to augment and automate intricate creative, analytical and optimisation routines. Traditional infrastructure engineering and CAD methods are still predominantly manual iterative simulation, integrating experts’ reasoning, heuristic improvement. Recent advances in LLMs, multi-agent systems and differentiable physics enable Autonomous Engineering Agents (AEAs) — AI agents effectively able to execute intention-driven design actions (i.e., goal-directed design behavior; geometry reasoning, negotiation of constraints in designs, natural language dialogue with engineers). As agents capable of self-regulation and as independent problem solvers that plan, design, analyze and refine engineered structures, these systems functions not only as aids to computation but active parts of the engineering loop. Towards the realisation of "self-driving CAD", in which design intent, physics, manufacturing limitations and optimisation aims are collaboratively recorded in a closed-loop multi-agent system, this work introduces both an inspired conceptual framework and corresponding technical frame for AEA. AEA s architectural elements are presented, comprising (1) a world model of geometry, material and simulation data, (2) specific agents for physics simulation, verification and documentation as well as for the generation of geometry and (3) an orchestrator that plans tracks and explains agent behavior. This architecture integrates differentiable geometry networks, LLM-driven planning and symbolic reasoning in a cohesive pipeline for human-in-the-loop (HITL) collaboration and modifiable autonomy. A prototype implementation demonstrates the autonomous refinemement of CAD models for structural components by combining finite element analysis (FEA), deep generative geometry, topology optimisation and automatic verification. The experimental setup provided a trade-off between executable designs, which could be verified for quality of design, and rapid reduction in human interaction. This shows that autonomous multi-level CAD refinement with respect to high level engineering goals is possible. Beyond mechanical design, AEA's broader vision is for civil and infrastructural systems in which autonomous agents can optimise routing, layout or safety compliance depending on evolving sustainability and regulatory targets. Finally, we discuss the socio-technical and ethical, as well as regulatory implications of AEAs emphasizing the need for human-in-the-loop, traceability and transparency. For responsible deployment in high-risk domains, we propose an open research agenda covering formal verification, explainability and certified surrogate models and policy frameworks. Engineering is progressing toward the future of intelligent, self-driving design ecosystems and AEAs are a revolutionary step toward AI systems that not only compute and visualize, but also think like designers, reason about designs, and collaborate on design creation.

References :

[1] Seshia, S. A., et al. “Design Automation for Cyber-Physical Systems: Challenges and Opportunities.” Proceedings of the IEEE, vol. 106, no. 9, 2018, pp. 1541–1562.

[2] Censi, A., et al. “Formal Methods for Cyber-Physical Systems: A Survey.” Annual Reviews in Control, 2020.

[3] Tang, M., and Zhao, Y. “Generative Design and Topology Optimization in Engineering.” ASME Journal of Computing and Information Science in Engineering, 2021.

[4] Bendsøe, M. P., and Sigmund, O. Topology Optimization: Theory, Methods, and Applications. Springer, 2019.

[5] Deisenroth, M. P., et al. “Gaussian Processes for Data-Efficient Learning in Robotics and Control.” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020.

[6] Kazi, R. H., et al. “Human-in-the-Loop Design: Enabling Interactive AI Systems.” Stanford HAI Reports, 2022.

[7] Ullman, T. D., et al. “Human–AI Collaboration in Design.” Nature Machine Intelligence, vol. 5, no. 3, 2023.

[8] Silver, D., et al. “Mastering the Game of Go with Deep Reinforcement Learning.” Nature, 2016.

[9] OpenAI. “GPT-4 Technical Report.” arXiv:2303.08774, 2023.

[10] Ha, D., et al. “Neural Representation of 3D Geometry for Design Automation.” arXiv preprint, 2022.

[11] Pan, Y., et al. “Mesh Autoencoders for Shape Representation.” Computer-Aided Design, 2021.

[12] Gao, W., et al. “Survey on 3D Shape Generation and Reconstruction.” Computer Graphics Forum, 2022.

[13] Reddy, P., and Banerjee, A. “Differentiable Simulation for Engineering Design.” Journal of Mechanical Design, 2021.

[14] Koch, C., et al. “Surrogate Models for Finite Element Analysis.” Structural and Multidisciplinary Optimization, 2020.

[15] Chen, Y., et al. “Hybrid Symbolic-Neural Architectures for Engineering Reasoning.” IEEE Transactions on Neural Networks and Learning Systems, 2022.

[16] Baldi, P., and Narayanan, R. “Knowledge Representation in Engineering AI.” AI for Engineering Design, Analysis and Manufacturing, 2020.

[17] Xie, Y., et al. “Multi-Agent Systems for Engineering Optimization.” Engineering Applications of Artificial Intelligence, 2021.

[18] Fan, Z., et al. “Collaborative Agents for Generative Product Design.” arXiv preprint, 2023.

[19] Ullrich, S., et al. “Formal Verification in Mechanical Design Systems.” Computer Methods in Applied Mechanics and Engineering, 2020.

[20] Deb, K., et al. Multi-Objective Optimization Using Evolutionary Algorithms. Wiley, 2018.

[21] Brookes, D., et al. “Causal Reasoning and Action Planning for Autonomous Systems.” AAAI Conference on Artificial Intelligence, 2023.

[22] McDermott, D., et al. “PDDL—The Planning Domain Definition Language.” Yale Technical Report, 1998.

[23] Duriez, C., et al. “Differentiable Physics Engines for Design.” ACM Transactions on Graphics, 2022.

[24] Sacks, E., and Joshi, S. “Computational Design Synthesis.” CIRP Annals, 2019.

[25] LeCun, Y., et al. “Self-Supervised Learning: The Dark Matter of Intelligence.” Nature Reviews Neuroscience, 2022.

[26] Russell, S. J., and Norvig, P. Artificial Intelligence: A Modern Approach. Pearson, 2021.

[27] Batty, M., et al. “Digital Twins and AI-Driven Design.” Philosophical Transactions of the Royal Society A, 2022.

[28] Wong, K., et al. “Trust and Explainability in Autonomous Engineering Systems.” IEEE Access, 2023.

[29] Zhao, L., et al. “Cyber-Physical Engineering with AI Agents.” Frontiers in Artificial Intelligence, 2024.

[30] Nguyen, T., et al. “From Design Automation to Autonomous Engineering Agents: A Roadmap.” arXiv preprint arXiv:2504.10234, 2025.

Keywords :

AI-driven design automation; Multi-Agent Systems; Generative Design; Human-in-the-Loop Engineering; Large Language Models; Differentiable Simulation; Infrastructure Automation; Design Verification; Autonomous engineering agents :Self-driving CAD.