Modes, Motions, and Morphology - Design Optimization for Legged Robots and Other Periodically Operating Mechanisms

• Applicant: Professor Dr. C. David Remy
• Subject Area: Mechanics ; Automation, Mechatronics, Control Systems, Intelligent Technical Systems, Robotics
• Term: since 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 501862165

The goal of this project is the development of a systematic design optimization method for mechanisms and machines that perform quasi-periodic motions. Think, for example, of a pick-and-place robot that selects objects from a crate and loads them into a production line. Even if such mechanisms don't perform substantial amounts of external work, they need energy to make up for mechanical losses due to damping, friction, and collisions. Most importantly, they must compensate negative actuator work that happens during the deceleration phase of reciprocal motions and that introduces the need for positive actuator work during the subsequent acceleration.A suitable design of the mechanical dynamics has the potential to greatly reduce the energy-need and increase the speed of such mechanisms. An interesting inspiration for such well-designed mechanical dynamics can be found in legged locomotion in nature. Animals have developed clever ways to reduce energy waste. In gaits, such as walking or running, a substantial part of the motion emerges passively from the mechanical structure of their bodies. Energy is not removed through negative actuator work, yet stored elastically in muscles and tendons or converted from kinetic to gravitational energy. Furthermore, motions are shaped by the pendulum dynamics of the body segments. By exploiting these dynamical effects, negative actuator work can be avoided and the necessary active contribution to the motion can be greatly reduced.The underlying premise is that the mechanical dynamics of such a system are able to passively create a substantial part of a desired motion which can hence be sustained with a minimal amount of effort. As the mechanical dynamics are a function of the system parameters (mass distributions, stiffness values, etc.) such an optimal exploitation inherently couples motion and morphology. Furthermore, for a complex system, the mechanical dynamics can normally be excited in a variety of discrete modes. In legged animals, these different modes manifest themselves as different gaits. Taking advantage of them, the natural dynamics can be exploited at more than one operating point; thereby allowing, for example, for efficient motion at different locomotion velocities. To become truly optimal, design and motion generation must thus hence be done at the same time.In this project, we will use advanced optimal control and machine learning to concurrently identify motions and morphologies that systematically exploit mechanical dynamics. By using different modes of the natural dynamics, we further seek to optimize a single design for multiple tasks.While the example application of this project will be the field of legged locomotion, the tools and methods that we develop in the process will be kept general, such that they can be applied to other domains of robotic or mechatronic systems. They will be packaged into a modular toolbox for the design optimization of periodically operating mechanisms.

• DFG Programme: Priority Programmes
• Subproject of: SPP 2353: Daring More Intelligence - Design Assistants in Mechanics and Dynamics