Computational Design of Robot Locomotion Through Large-Scale Root-Finding

University Of California-Berkeley, Berkeley CA

Investigators

Abstract

The ability of small legged robots to move in complicated terrain, such as a disaster zone, depends on the leg geometry. For example, robot legs might be desired to climb obstacles as well as to run quickly. Currently, the design of mechanical limbs for robot locomotion is primarily performed through intuition or by incrementally improving pre-existing designs, rather than from a systematic approach. This award supports fundamental research into the creation of a design procedure for robot legs for a given locomotion task. This exploration is enabled by a new algorithm for solving polynomial equations that substantially reduces computational requirements to calculate necessary leg geometry. One outcome of success in this research project will be design tools for small low-cost robots with exceptional mobility, which could be distributed throughout a disaster zone to locate survivors. Code developed as part of this project will be made publically available, and will allow non-experts, such as undergraduates, the ability to design sophisticated linkages for new applications and products. This research project aims to formulate a new robot design method where locomotive requirements are encoded into polynomial systems in which the roots to these systems represent potential designs for robot limbs. Central to the design algorithm is a new polynomial root-finding method termed Finite Root Generation (FRG). Initial estimates reveal the number of roots to these problems is impractically large, but there is good evidence that the true number of roots, called finite roots, is smaller by several orders of magnitude. Finite roots are the only type of root with engineering relevance. Therefore, the FRG method is employed to handle the polynomial systems of interest in this project. This technique offers large computational savings as it finds only finite roots to a polynomial system, and is able to estimate the size of a solution set without computing the entire set. This project formulates a new design approach that considers locomotive quantities, encodes requirements into design equations, and obtains useful design solutions, which all together form a powerful design exploration toolset for robot locomotion platforms.