COSMO-Agent: Tool-Augmented Agent for Closed-loop Optimization, Simulation, and Modeling Orchestration
Summary: arXiv:2604.05547v1 Announce Type: new
Abstract: Iterative industrial design-simulation optimization is bottlenecked by the CAD-CAE semantic gap: translating simulation feedback into valid geometric edits under diverse, coupled constraints. To fill this gap, we propose COSMO-Agent (Closed-loop Optimization, Simulation, and Modeling Orchestration), a tool-augmented reinforcement learning (RL) framework that teaches LLMs to complete the closed-loop CAD-CAE process.
Specifically, we cast CAD generation, CAE solving, result parsing, and geometry revision as an interactive RL environment, where an LLM learns to orchestrate external tools and revise parametric geometries until constraints are satisfied. To make this learning stable and industrially usable, we design a multi-constraint reward that jointly encourages feasibility, toolchain robustness, and structured output validity. In addition, we contribute an industry-aligned dataset that covers 25 component categories with executable CAD-CAE tasks to support realistic training and evaluation. Experiments show that COSMO-Agent training substantially improves small open-source LLMs for constraint-driven design, exceeding large open-source and strong closed-source models in feasibility, efficiency, and stability.
Introduction
The integration of Computer-Aided Design (CAD) and Computer-Aided Engineering (CAE) has long been a challenge in industrial design. The semantic gap between these two domains creates barriers in translating simulation feedback into actionable geometric changes. COSMO-Agent aims to bridge this gap by utilizing advanced reinforcement learning techniques.
Key Features of COSMO-Agent
- Tool-Augmented Framework: COSMO-Agent employs a reinforcement learning approach that integrates external tools necessary for CAD and CAE processes.
- Interactive Learning Environment: The environment allows the Large Language Model (LLM) to learn through interaction, enabling it to orchestrate tools and refine designs effectively.
- Multi-Constraint Reward System: This innovative reward system fosters the development of feasible designs while ensuring the robustness of the toolchain and the validity of outputs.
- Industry-Aligned Dataset: The dataset features 25 component categories and executable tasks, ensuring that the training is relevant and applicable to real-world scenarios.
Impact on Industrial Design
COSMO-Agent represents a significant advancement in the field of industrial design, particularly in the realm of constraint-driven design. By facilitating a smoother closed-loop process, it enhances the capacity for rapid iteration and optimization, which is crucial in today’s fast-paced industrial environment.
Experimental Results
Initial experiments demonstrate that COSMO-Agent training results in notable improvements in performance metrics. The system has shown superior feasibility and efficiency compared to both large open-source models and leading closed-source alternatives. This positions COSMO-Agent as a formidable tool for engineers and designers seeking to optimize their design processes.
Conclusion
The introduction of COSMO-Agent marks a pivotal step in addressing the challenges posed by the CAD-CAE semantic gap. By leveraging reinforcement learning, it not only enhances the design-simulation optimization process but also sets a new standard for future developments in this area. As industries continue to evolve, tools like COSMO-Agent will be essential in driving innovation and efficiency.