Sven Cremer

ORCID Identifier(s)

0000-0002-8889-0508

Graduation Semester and Year

2017

Language

English

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Electrical Engineering

Department

Electrical Engineering

First Advisor

Dan O Popa

Abstract

With an increasing number of collaborative robots or “co-robots” entering human environments, there is a growing need for safe, intuitive, and efficient (physical) Human-Machine Interfaces. Unlike industrial robots, co-robots operate in cluttered and dynamic working spaces, where accidental collisions are more likely to occur. To minimize interaction forces, co-robots are usually lightweight and compliant. However, this makes the robot dynamics highly nonlinear and therefore difficult to model and control. In addition, the control loop must incorporate feedback from integrated sensors. Future systems under development are covered with force-sensing robot skin comprised of thousands of multi-modal sensors, creating the need for efficient robot sensor calibration and processing. During this thesis work, a neuroadaptive (NA) controller was developed and validated for safe and stable physical interaction. In order to achieve intuitive physical Human-Robot Interaction (pHRI), the robot error dynamics were modified to behave equivalent to a simple admittance model by expanding the NA controller with prescribed error dynamics (PED). Another new approach for modifying the robot error dynamics was an inner/outer-loop structure consisting of an admittance model in the outer-loop, which generates a model trajectory that the inner-loop follows. This admittance model was implemented with an autoregressive moving average (ARMA) filter, which was tuned with recursive least squares and with the help of a prescribed task model. Experiments conducted during this thesis showed that the developed two-loop framework allows a high degree of generality and adaptability to different human preferences, tasks, robots, and sensors. It is also offers a novel algorithm for adaptive calibration of robot skins by directly tuning admittance models that map sensor voltages into desired robot motion. Finally, it was suggested that the pHRI can be made more efficient by reducing the human effort during a collaborative task. The human force exerted on the robot to achieve a desired pose can be minimized by predicting and then executing the desired human motion. Different human intent estimators (HIEs) were proposed, including a neural network based estimator.

Keywords

co-robots, HMI, pHRI, neuroadaptive, adaptive control, robot skin, calibration, human intent estimation, SkinSim, SkinLearn

Disciplines

Electrical and Computer Engineering | Engineering

Comments

Degree granted by The University of Texas at Arlington

Download

Share

COinS