Research, Legged Locomotion, Manipulation

Overview

This is an invited research paper I wrote in collaboration with the MARMot Lab at the National University of Singapore, and the Daltorio Lab at Case Western Reserve University, for the journal Frontiers in Mechanical Engineering.

Abstract

Legged robots can have a unique role in manipulating objects in dynamic, humancentric, or otherwise inaccessible environments. Although most legged robotics research to date typically focuses on traversing these challenging environments, many legged platform demonstrations have also included “moving an object” as a way of doing tangible work. Legged robots can be designed to manipulate a particular type of object (e.g., a cardboard box, a soccer ball, or a larger piece of furniture), by themselves or collaboratively. The objective of this review is to collect and learn from these examples, to both organize the work done so far in the community and highlight interesting open avenues for future work. This review categorizes existing works into four main manipulation methods: object interactions without grasping, manipulation with walking legs, dedicated nonlocomotive arms, and legged teams. Each method has different design and autonomy features, which are illustrated by available examples in the literature. Based on a few simplifying assumptions, we further provide quantitative comparisons for the range of possible relative sizes of the manipulated object with respect to the robot. Taken together, these examples suggest new directions for research in legged robot manipulation, such as multifunctional limbs, terrain modeling, or learning-based control, to support a number of new deployments in challenging indoor/outdoor scenarios in warehouses/construction sites, preserved natural areas, and especially for home robotics.

Personal Motivation

While writing my bachelor’s thesis as a visiting researcher at MARMot Lab, I undertook various additional projects exploring legged robots. Some of these entailed investigating their ability to perform useful manipulation tasks with minimal augmentation to their mechanical design. Since there are many possibile ways of implementing such an idea, I searched for a coherent account for this manipulation technique and was disappointed when I was unable to find one. Since our labs consisted of people with individual expertise in various segments of this topic, we decided to put our heads together and create such a resource by organizing the community’s existing body of work, and highlighting interesting avenues for future research.

Acknowledgements

This was a product of the combined effort of Yifeng Gong, Ge Sun, Aditya Bidwai, Raghuram, and myself. It was truly a pleasure to work with them, and I also learned a lot about writing meaningful research in my first such endeavour. The support and invaluable advice from our advisors: Dr. Guillaume Sartoretti and Dr. Kathryn Daltorio was paramount in making this a great experience. Thanks Guillaume, for seeing the potential in my writing :)