AI Agents in Engineering Design - Multi-Agent Framework for Car Design

Revolutionizing Car Design with AI

From weeks to minutes: Our multi-agent AI framework accelerates the entire automotive design process by seamlessly integrating conceptual sketching, 3D modeling, CFD meshing, and aerodynamic simulations.

Styling Agent

Transforms hand-drawn sketches into high-resolution, photorealistic renderings using SDXL and ControlNet

CAD Agent

Retrieves similar 3D designs from DrivAerNet++ and generates new shapes using DeepSDF

Meshing Agent

Automatically generates high-quality CFD meshes using OpenFOAM's snappyHexMesh

Simulation Agent

Provides real-time aerodynamic predictions using TripNet surrogate models

Abstract

We introduce the concept of "Design Agents" for engineering applications, particularly focusing on the automotive design process. Our framework integrates AI-driven design agents into the traditional engineering workflow, demonstrating how these specialized computational agents interact seamlessly with engineers and designers to augment creativity, enhance efficiency, and significantly accelerate the overall design cycle. By automating and streamlining tasks traditionally performed manually, such as conceptual sketching, styling enhancements, 3D shape retrieval and generative modeling, computational fluid dynamics (CFD) meshing, and aerodynamic simulations, our approach reduces certain aspects of the conventional workflow from weeks and days down to minutes.

Key Innovations

Multi-Agent Orchestration

Leverages AutoGen framework to coordinate specialized AI agents, enabling seamless collaboration between different design tasks and maintaining context throughout the workflow.

Rapid Design Iteration

Reduces traditional design cycles from weeks to minutes by automating sketch refinement, 3D generation, meshing, and simulation processes.

Cross-Modal Intelligence

Integrates vision-language models (VLMs), large language models (LLMs), and geometric deep learning to bridge 2D concepts with 3D engineering requirements.

DrivAerNet++ Integration

Utilizes the largest multimodal car dataset with 8,000 designs and high-fidelity CFD simulations for training and validation.

Real-time Feedback

Provides instant aerodynamic predictions using surrogate models, enabling designers to immediately assess performance implications.

Human-AI Collaboration

Augments rather than replaces human creativity, serving as intelligent assistants that enhance the design process.

Agent Workflow

1. Conceptual Sketching

Designer creates a hand-drawn sketch of the car concept, capturing initial aesthetic vision and basic proportions.

2. Styling Enhancement

Styling Agent transforms sketches into photorealistic renderings using Stable Diffusion XL with ControlNet guidance.

3. 3D Shape Generation

CAD Agent retrieves similar designs from DrivAerNet++ or generates new 3D geometries using DeepSDF interpolation.

4. Mesh Generation

Meshing Agent automatically creates CFD-ready meshes using OpenFOAM's snappyHexMesh with quality verification.

5. Aerodynamic Analysis

Simulation Agent provides real-time predictions of drag coefficient and flow patterns using TripNet surrogate models.

Results & Demonstrations

Figure 1

AI Design Agents Framework for Automotive Design

Overview of our multi-agent system integrating foundation models (LLMs and VLMs), geometric deep learning models, and engineering tools. The framework shows how four specialized agents—CAD Agent, Styling Agent, Meshing Agent, and Simulation Agent—collaborate to accelerate the car design process from initial sketch to aerodynamic analysis.

Figure 2

Styling Agent: From Sketches to Photorealistic Renderings

Demonstration of the Styling Agent's capability to transform hand-drawn sketches into diverse, high-resolution car renderings using Stable Diffusion XL with ControlNet. The agent generates various aesthetic styles including modern sedans, luxury vehicles, sports cars, vintage designs, and rugged SUVs, each with distinctive visual characteristics and color schemes based on text prompts.

Figure 3

Real-time Styling Process

Animation showing the iterative refinement process of the Styling Agent, demonstrating how sketches are progressively enhanced with different styles, materials, and lighting effects.

Figure 3a

3D Shape Interpolation (Bottom View)

DeepSDF-based interpolation between fastback (detailed/smooth), estateback, and notchback configurations showing underbody variations.

Figure 3b

3D Shape Interpolation (Front View)

Front perspective of shape morphing demonstrating smooth transitions in the learned latent space for automotive geometries.

Figure 4

Real-time Aerodynamic Analysis via Simulation Agent

Comparative aerodynamic analysis showing velocity field visualizations and drag coefficients for 15 different car designs from the DrivAerNet++ dataset. Each design shows the car ID, drag coefficient (Cd), relative performance change, and corresponding CFD simulation results. The Simulation Agent retrieves these pre-computed high-fidelity simulations in real-time, enabling immediate performance assessment during the design process.

Figure 5

End-to-End Design Workflow

Complete demonstration of the multi-agent system in action, showing the seamless transition from initial sketch through styling, 3D generation, meshing, to final aerodynamic analysis—reducing the traditional weeks-long process to minutes.

AI Agents in Engineering Design: A Multi-Agent Framework for Aesthetic and Aerodynamic Car Design