Co-Design Assistants for Mechatronic Systems: Enabling the Sim-to-Real Transfer

Overall goal: Development of a co-design assistant for mechatronic systems that jointly optimizes hardware and control, explicitly accounting for their strong interdependence throughout the design process.

• Address sim-to-real transfer challenges by designing for robustness, such that discrepancies between simulation and physical systems can be compensated primarily through control adaptation rather than hardware modification.
• Employ surrogate models that combine first-principles modeling with data-driven techniques to enable efficient and reliable co-design optimization.
• Support gradient-based optimization through models that provide derivative information and simplify the design problem via convexification or latent-space representations enabled by modern machine learning.
• Integrate experimental real-world data into iterative design loops to reduce model mismatch and improve transferability.
• Recognize the asymmetry between fixed hardware and adaptable control by prioritizing hardware choices that remain robust across a wide range of control strategies.
• Validate the proposed co-design assistant on three representative robotic systems, demonstrating improved performance and sim-to-real reliability.