Project Details
Projekt
Co-Design Assistants for Mechatronic Systems: Enabling the Sim-to-Real Transfer
Subject Area
Automation, Mechatronics, Control Systems, Intelligent Technical Systems, Robotics
Mechanics
Mechanics
Term
since 2025
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 564445282
The co-design of mechatronic systems entails the simultaneous and integrated development of physical components (such as actuators and mechanical structures) and control algorithms. This approach acknowledges the profound interdependence between hardware and software in mechatronic systems, where changes in one can significantly impact the performance of the other. For example, optimizing the design of a robotic leg's joints and actuators must go hand in hand with refining motion planning and control algorithms to achieve both speed and energy efficiency. By treating hardware and software as components of a unified system rather than separate entities, co-design enables the development of mechatronic systems that are more efficient, adaptive, and capable of handling complex tasks. Model-based optimization is a powerful tool for tackling co-design challenges: It employs objective functions to capture desired performance metrics, such as efficiency or speed, while incorporating constraints that represent physical limitations, stability requirements, or operational bounds. For instance, in the co-design of a legged robotic system, optimization can simultaneously adjust hardware specifications (e.g., mass distribution or joint stiffness) and control parameters (e.g., motion trajectories or feedback gains) to maximize speed or minimize energy consumption. However, significant differences between these variables arise when transferring optimization results from simulation to real-world implementation: Changing hardware specifications is costly and time-consuming, while the mechatronic system can utilize a variety of motion patterns, each governed by control parameters that can be easily adapted post-implementation. Data on hardware variable variations is typically sparse, while control variables can be explored extensively. This inherent gap between simulation and physical realization necessitates a specific type of robustness: Systems must be designed to address discrepancies between models and hardware by relying solely on adjustments to control parameters—ideally without compromising performance. The goal of this project is to develop a design assistant tailored for mechatronic co-design that addresses these challenges. The assistant will leverage numerical optimization on surrogate models that combine first-principle modeling with data-driven techniques and are specifically designed for this purpose. These models will streamline optimization and control by providing gradient and derivative information while simplifying the problem—using methods like convexification or mapping to higher-dimensional spaces enabled by modern machine learning approaches. Experimental real-world data will be integrated to support iterative design processes. A primary focus will be ensuring reliable transfer of optimization results from simulation to physical implementation. The proposed approach will be validated on three exemplary robotic systems.
DFG Programme
Priority Programmes
