Whole-Body Control of Humanoid Robots

Abstract

Whole-body controllers have become a well-adopted control paradigm for humanoid robots, uniting communities working on control systems and sensor-based techniques. These communities have focused on developing simple design tools and adopting common math representations. As such, we are now closer than ever toward sharing common techniques that work across platforms and making the process of designing controllers as easy as possible, with the hope of one day to completely automate the controller design process for all types of humanoid robots. At the same time, current whole-body controllers suffer from important limitations that prevent them from achieving highly agile dynamic behaviors, operating with high mechanical and electronic efficiency, be safe at high speeds in terms of human-robot interactions, and be easy to deploy with minimal designer tuning. This chapter aims to explore various key issues in whole-body control of humanoid robots, including (i) providing an in-depth overview of state-of-the-art techniques, (ii) providing a list of references to open-source software implementations, (iii) discussing performance differences between the various types of architectures, and (iv) analyzing limitations and future research directions for optimal performance.