Engineering Compliance in Legged Robots Via Robust Co-Design

Abstract

This article presents a design framework for the scalable co-design of hardware and control as applied to improving the energy efficiency of legged robots with parallel compliance. The proposed framework uses the Alternating Direction Method of Multipliers for design synthesis by solving large-scale trajectory optimization problems. Specifically, we use Stochastic Programming constructs to model design uncertainty associated with terrain properties, and enforce robustness by co-optimizing the robot morphology, a nominal trajectory, and a feedback control policy. Our framework is applied to tune the design of parallel elastic actuation (PEA) via considering how the PEA can be used to actively tailor compliance to different locomotion scenarios. The design optimization framework is validated with the MIT Mini Cheetah quadruped, where added compliance reduces its cost of transport by 58.3% in simulation of optimized planar bounding gaits, and up to 17.4% and 8.3% in experiments when executing trotting and bounding gaits, respectively.