Abstract

A group of individuals can coordinate very easily to carry a large or heavy object together. This act of collaborative carrying is called co-manipulation. Unlike human teammates, robots cannot easily detect and respond to the intent of human partners during co-manipulation tasks. This shortcoming can be compensated for via the physical design and the software design of robots. This work treats software design with a neural net that detects human modus (or the manner in which a team moves an object together) in real time. This work treats physical design with the use of a soft robot in human-robot co-manipulation, and a simple control system which leverages its inherent compliance. Two studies were performed during this work. The first study adapted recent work from the BYU Robotics And Dynamics Lab (RaD lab) by Seth Freeman. His work was a neural net co-manipulation classifier and the first objective of this thesis was to adapt it to work in real time. The study consisted of 5 pairs of people carrying a 1.2m long, 22.5kg object through a physical obstacle course while the aforementioned neural net classifier classified their intended behavior as either moving quickly, smoothly, or avoiding obstacles. The real-time classifier was correct 79.6% of the time and was shown to be robust to circumstances beyond its original training data. The second study builds on the recent work of Shaw and Jackson, also from the BYU RaD Lab, and consisted of 25 individuals who each carried a 11kg, 1.2m object through a series of tasks with the assistance of a soft robot partner. Rather than a physical obstacle course, this second study used virtual reality (VR) to guide participants through tasks. Throughout this second study, various control parameters of the robot were changed and performance was analyzed with quantitative metrics as well as surveys given to the participants about preference and performance. The second study revealed that more compliant settings for the soft arm were performed slightly better than less compliant settings, but that people's ability to adapt and improve performance with experience outweighs the initial difference in performance due to distinct stiffnesses. Additionally, these results were compared to results from Shaw and Jackson's human-human co-manipulation study and it was found that the robot has much to improve to match human performance, but is still a good benchmark.

Degree

MS

College and Department

Ira A. Fulton College of Engineering; Mechanical Engineering

Rights

https://lib.byu.edu/about/copyright/

Date Submitted

2025-04-16

Document Type

Thesis

Handle

http://hdl.lib.byu.edu/1877/etd13566

Keywords

physical human-robot interaction, co-manipulation, soft-robotics

Language

english

Download

Included in

Engineering Commons

Share

COinS