Co-Design of Robotic Mechanisms

Graphic showing the SP co-design process for a robotic manipulator and a monopod robot
SP co-design process for a robotic manipulator and a monopod robot (“One robot for many tasks: Versatile co-design through stochastic programming,” ICRA 2020).

Overview

A co-design framework manages and solves two or more design problems simultaneously rather than consecutively. Design simultaneity is particularly important when dealing with complex systems, such as those included in robotics mechanisms. Concurrency in a design dictates the fulfillment of constraints and functionality requirements shared among two or more design problems. The two main design problems in robotics involve control design and morphology design. The synergy between morphology (body) and control (brain) would enable a robot to reach mobility and skill levels comparable to those seen in animals.

Technical Approaches

Although the emulation of biological systems can provide a direction to achieve better designs, having the morphology decoupled from the control leaves a robot’s performance entirely reliant on its controllers. As a result, the controllers can become overload and unreliable if the environment's conditions deviate from those for which the control was designed and tuned. To overcome such challenges, we propose to equip the co-design process with optimization tools and methods to establish connections between typical design constraints while minimizing the resources required to achieve specific functionalities.

We use trajectory optimization as a foundation to state and solve co-design problems expressed as optimal control problems. Furthermore, we expand the design space by modeling uncertainty at design time when considering multiple scenarios that a robot could face. To deal with uncertainty at design time, we use stochastic programming (SP) as a strategy to span the performance of a robot over many and varied conditions (Figure). We envision applying our co-design framework to produce robots capable of exploiting feedback pathways encoded in their morphology, reducing the intervention of control and sensory systems. In achieving our goal, we will build a co-design approach that is more versatile and closer to nature in the efficient distribution of skills and resources.