Design principles of multi-axis, large magnitude force sensors based on stress fields for use in human and robotic locomotion

• Main Author: Chuah, Meng Yee (Meng Yee Michael)
• Other Authors: Sangbae Kim.
• Format: Others
• Language: English
• Published: Massachusetts Institute of Technology 2018
• Subjects: Mechanical Engineering.
• Online Access: http://hdl.handle.net/1721.1/119276

Thesis: Ph. D., Massachusetts Institute of Technology, Department of Mechanical Engineering, 2018. === This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections. === Cataloged from student-submitted PDF version of thesis. === Includes bibliographical references (pages 151-163). === Our ability to purposefully move across varied terrain requires us to have knowledge of the interactions our feet have with the external environment. However, existing sensing methods are inadequate to address the many unique demands of legged locomotion (i.e. fragile structures, incapable of handling large impact forces and noise caused by inertial loads during stride). This research is a study of how best to replicate the role of skin mechanoreceptors that enable our biological counterparts to perform dynamic maneuvers, and to develop innovative sensors that would empower the next generation of agile robots and smart shoes. The thesis introduces new design principles and methodologies for developing multi-axis, large magnitude force sensors based on stress fields to achieve these goals. Fabrication methods are presented for a monolithic elastomeric footpad that is biologically inspired, allowing it to measure large magnitude forces in both normal and shear axes while being compact, lightweight, impact robust, dust tight, and waterproof. The key principle that enables this is termed Stress Field (SF) based force sensing. Instead of funneling the load path directly through a few sensors in traditional force sensing methods, SF based force sensing allows the sampling of the stress distribution over the entire footpad surface through an array of piezoresistive sensor elements. The force estimator is constructed in two steps. First, linear regression fits the sensor readings to normal and shear forces. Then, machine learning is used as a nonlinear function approximator on the residual to further refine the force estimator to achieve greater accuracy. To enable these SF force sensor to be reproduced or customized for different needs, guidelines are provided in the form of simple design principles based on biological receptive fields, as well as an analytical model for cylindrical sensor types. For more complex sensor geometries, a material model of the elastomer is experimentally characterized, and Finite Element Analysis (FEA) can be used to determine the optimal configurations of these sensor arrays for different sensing needs. To show the feasibility of these SF force sensors, they have been validated for both robotic and human locomotion. For robotic locomotion, a hemispherical design was developed and implemented on the MIT Cheetah, a quadrupedal running robot, as well as on Little HERMES, a bipedal robot. For human locomotion, two prototypes of force sensing shoes have been fabricated based on cylindrical SF force sensors as a proof of concept. In the future, these lightweight, low-cost, multi-axis force sensors can be customized for different applications and fully integrated into smart shoes, prosthetic devices, and robotic exoskeletons to provide the real-time ground reaction force data. This data would enable new capabilities in various fields such as healthcare, sports analytics, virtual reality, and robotics. === by Meng Yee (Michael) Chuah. === Ph. D.