Robust Analysis for Mechanism and Behavior Co-optimization of High-performance Legged Robots

Abstract

This paper presents a novel application of robust analysis on the mechanism and behavior co-optimization of a virtual high-performance monopedal robot. The analysis is tailored for the design of high-performance legged robots, and is based on realistic models with a large number of parameters, that are designed to achieve a large variety of tasks. Robust analysis takes into consideration the effects that real imperfections can have on the performance of the real design; something which is neglected in most design studies and can be crucial for certain applications. The results demonstrate that typical optimization approaches tend to over-optimize the model, which results in a theoretical performance that is practically impossible to achieve in reality. Instead, the proposed methodology takes into consideration imperfections of the real world to obtain an average performance that is less sensitive to these uncertainties, and can lead to robot designs that are closer to their design expectations.